U.S. patent application number 10/117356 was filed with the patent office on 2008-06-05 for system to determine quality through reselling of items.
This patent application is currently assigned to Emergent Music LLC. Invention is credited to Gary Robinson.
Application Number | 20080133417 10/117356 |
Document ID | / |
Family ID | 46328237 |
Filed Date | 2008-06-05 |
United States Patent
Application |
20080133417 |
Kind Code |
A1 |
Robinson; Gary |
June 5, 2008 |
System to determine quality through reselling of items
Abstract
A system for determining a market value of items includes
establishment of a sequence of items to be sold or licensed and a
sale of income rights to the sale or licensure. A table of income
rights corresponding to the sale to consumers is established, and
speculators are able to make determinations of the potential for
future sales of the items. After sale or licensure to consumers,
the transactions to consumers are considered finalized, and
speculators receive income based on the transfer.
Inventors: |
Robinson; Gary; (Bangor,
ME) |
Correspondence
Address: |
ELMAN TECHNOLOGY LAW, P.C.
P. O. BOX 209
SWARTHMORE
PA
19081
UNITED STATES
610-892-9942
925-226-4995
GERRY@ELMAN.COM
|
Assignee: |
Emergent Music LLC
565 Congress Street #201
Portland
ME
04101
|
Family ID: |
46328237 |
Appl. No.: |
10/117356 |
Filed: |
April 5, 2002 |
Related U.S. Patent Documents
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
|
|
09/691,316 |
Oct 18, 2000 |
|
|
|
10/117,356 |
|
|
|
|
60/160,044 |
Oct 18, 1999 |
|
|
|
60/166,430 |
Nov 19, 1999 |
|
|
|
60/165,794 |
Nov 16, 1999 |
|
|
|
60/176,154 |
Jan 14, 2000 |
|
|
|
60/176,953 |
Jan 18, 2000 |
|
|
|
60/182,836 |
Feb 16, 2000 |
|
|
|
60/194,988 |
Apr 5, 2000 |
|
|
|
60/200,204 |
Apr 28, 2000 |
|
|
|
60/209,930 |
Jun 7, 2000 |
|
|
|
60/218,866 |
Jul 18, 2000 |
|
|
|
60/232,742 |
Sep 15, 2000 |
|
|
|
60/281,673 |
Apr 5, 2001 |
|
|
|
Current U.S.
Class: |
705/52 |
Current CPC
Class: |
G06Q 30/0282 20130101;
G06Q 30/00 20130101 |
Class at
Publication: |
705/052 |
International
Class: |
H04L 9/00 20060101
H04L009/00; H04K 1/00 20060101 H04K001/00 |
Claims
1. A system for valuation and financing of items, the system
comprising: determining a sales unit of an item to be financed;
establishing financial rights for the sales units, by which
transfer of the sales unit is conditional on at least a partial
transfer of the financial rights for that sales unit; segmenting
groupings of the financial rights according to a substantially
sequential production of the sales units; selling the financial
rights.
2. The system of claim 1, wherein the sales unit is a copyright
license.
3. The system of claim 1, wherein the financial right accrues to
the buyer, so that upon transfer of at least a portion of the sales
unit, the buyer of the financial rights becomes entitled to a
return from the financial rights.
4. The system of claim 1, wherein: the sales unit is a copyright
license and the financial rights are rights to royalties from the
copyright license; the rights to royalties accrue to the buyer, so
that upon transfer of at least a portion of the copyright license,
the buyer of the rights to royalties becomes entitled to a return
from the rights to royalties.
5. The system of claim 1, further comprising: determining an
initial production run; determining an incremental production run;
and establishing the financial rights based on the incremental
production run.
6. The system of claim 1, further comprising any two or more of an
Input Unit, a Combining Unit, an Input Valuation Unit, and a Reward
Unit.
7. A method for financing of items, the method comprising:
determining a sales unit of an item to be financed; establishing
financial rights for the sales units, by which transfer of the
sales unit is conditional on at least a partial transfer of the
financial rights for that sales unit; segmenting groupings of the
financial rights according to a substantially sequential production
of the sales units; selling the financial rights.
8. The method of claim 7, wherein the sales unit is a copyright
license.
9. The method of claim 7, wherein the financial right accrues to
the buyer, so that upon transfer of at least a portion of the sales
unit, the buyer of the financial rights becomes entitled to a
return from the financial rights.
10. The method of claim 7, wherein: the sales unit is a copyright
license and the financial rights are rights to royalties from the
copyright license; the rights to royalties accrue to the buyer, so
that upon transfer of at least a portion of the copyright license,
the buyer of the rights to royalties becomes entitled to a return
from the rights to royalties.
11. The method of claim 7, further comprising: determining an
initial production run; determining an incremental production run;
and establishing the financial rights based on the incremental
production run.
12. The method of claim 7, further comprising any two or more of an
Input Unit, a Combining Unit, an Input Valuation Unit, and a Reward
Unit.
Description
CROSS-REFERENCE TO RELATED APPLICATIONS
[0001] This application claims priority from U.S. Provisional
Patent Application No. 60/281,673, filed Apr. 5, 2001. The entire
contents of the aforementioned patent application are incorporated
herein by reference.
[0002] This application claims priority as a continuation-in-part
application from U.S. patent application Ser. No. 09/691,316,
filed, Oct. 18, 2000. The entire contents of the aforementioned
patent application are reproduced below:
IMPROVED ONLINE COMMUNITY THROUGH RATINGS
Cross References to Related Applications
United States Provisional Patent Applications
[0003] Community-Based Market Movement Prediction: No. 60/160,044;
filed Oct. 18, 1999 [0004] Portfolio Management By Community; No.
60/166,430; filed Nov. 19, 1999 [0005] Clusters for Rapid
Artist-Audience Matching: No. 60/165,794; filed Nov. 16, 1999
[0006] A Mechanism for Quickly Identifying High-Quality Items: No.
60/176,154; filed Jan. 14, 2000 [0007] A Mechanism for Quickly
Identifying High-Quality Items Version 000118: No. 60/176,953,
filed Jan. 18, 2000 [0008] A Mechanism for Quickly Identifying
High-Quality Items Version 000216; No. 60/182,836, filed Feb. 16,
2000 [0009] A Mechanism for Quickly Identifying High-Quality Items
Version 000405; No. 60/194,988, filed Apr. 5, 2000 [0010] A
Mechanism for Quickly Identifying High-Quality Items: No.
60/200,204; filed Apr. 28, 2000 [0011] A Mechanism for Quickly
Identifying High-Quality Items: No. 60/209,930; filed Jun. 7, 2000
[0012] A Mechanism for Quickly Identifying High-Quality Items: No.
60/218,866; filed Jul. 18, 2000 [0013] A Mechanism for Quickly
Identifying High-Quality Items: No. 60/232,742 filed Sep. 15,
2000
[0014] The entire disclosures thereof of the above-enumerated
United States Provisional patent applications, including the
specifications, drawings, claims and abstracts, are considered as
being part of the disclosure of this application and are hereby
incorporated by reference herein.
BRIEF SUMMARY OF THE INVENTION
[0015] The invention involves some or all of an Input Unit, a
Combining Unit, an Input Valuation Unit, and a Reward Unit (this
list is not meant to be inclusive).
[0016] The Input Unit is a facility or medium in which users can
register opinions. A non-inclusive, list of workable Input Units
is: an electronic listing of musical recordings with attached means
for the input of ratings of those recordings, an electronic listing
of securities with input means for predictions of price movements
of those securities, and "message board" or "discussion group"
software in which messages can be rated by the readers of those
messages.
[0017] The Combining Unit is a calculating mechanism accepting
input received from a plurality users and combining it into a
single value. In various embodiments, arithmetic averaging,
geometric averaging, meta-analytical, or Bayesian calculations,
among other possibilities, are used. Bayesian expectations based
upon the beta or Direchlet distributions are among the applicable
Bayesian calculations. Meta-analytical calculations include the
inverse normal method and Fisher's inverse chi-square method. The
Combining Unit is used to detect an overall group opinion, such as
stock market prediction or rating of a message, in the input given
by users regarding an item.
[0018] When using the inverse normal meta-analytical method, a
correlation matrix between the various users' ratings can be
calculated and used in calculating the combined p-value.
[0019] The Input Valuation Unit has responsibility for determining
the value of each user's input to the system.
[0020] The key values involved in this calculation are the
accuracy, timeliness, and volume of the user's ratings. However
various embodiments may also incorporate to other values, or leave
one or more of the mentioned values out of the calculations.
[0021] Some embodiments do this by assigning a value to every
rating individually, then summarizing this information to create an
assessment of the user who generated those ratings. Some
embodiments never create an individual value for each rating, but
rather use the ratings supplied by each user in to calculate an
overall valuation of the individual's contributions.
[0022] The key attributes which are considered by the calculations
for the value of individual ratings are their timeliness and
accuracy.
[0023] In various embodiments, accuracy measurements for individual
ratings are based on the difference between the individual's rating
and the population average, the absolute value of that difference,
the square of that difference, and other calculations that are
representative of what is intuitively understood as the general
degree of nonsimilarity between two values. In some such
embodiments the individual's rating is ignored in counting the
population average. In some others all ratings supplied earlier
than the one in question are ignored due to their possible
influence on the mindset of the rater.
[0024] Accuracy measurements for the individual's overall
contribution may be created by first calculating accuracy for the
individual ratings and then calculating their (arithmetic or
geometric) average or a Bayesian expectation or a z-score of
statistical significance calculation for any tendency toward
accurate or inaccurate ratings, or, in still further embodiments,
other approaches are used. Alternatively, no measure of the
individual value of each rating is calculated, and instead, an
overall measure such as statistical correlation is used. When
correlation is used, the correlation is measured between the user's
ratings and the population averages, which, in various embodiments,
may exclude the user's ratings and/or all previous ratings. Some
other embodiments use such techniques as normalizing the difference
between the individual rating and the population to be a real
number between 0 and 1 which is uniformly distributed under the
null hypothesis of randomness; such normalization techniques enable
further statistical calculations to be carried out, such as
combining these differences (which can now be considered to be
p-values) using meta-analytical calculations including the inverse
normal technique and Fisher's inverse chi-square technique. Some
other embodiments calculate a Bayesian expectation of these
differences, where the difference are or are not normalized,
according to the particular embodiment. This list, like other such
lists in the summary, however, is for purposes of example only and
is not meant to be inclusive.
[0025] Where population averages are mentioned in the above
paragraphs, some implementations use a Bayesian expectation of the
next rating generated by the population rather than use the average
itself.
[0026] When Bayesian expectations are mentioned, various
embodiments base such calculations on the beta and Direchlet
distributions, and other distributions.
[0027] In some embodiments which work in conjunction with external
markets such as stock prices on the New York Stock Exchange,
accuracy is determined by comparing the predictions of users with
the valuations that eventually emerge in the market. A prediction
that a stock will have a certain price at a future date and time
can be considered a rating of the stock. If that rating differs
from the current price, the rater is saying that he disagrees with
the overall ratings currently provided by the community as a whole,
but thinks that the community will agree with him at the future
date and time. Thus some of accuracy calculations described above,
such as the correlation between the user's rating and the rating
the community as a whole, are calculated in various such
embodiments.
[0028] Similarly, in embodiments where the items being rated are
Web pages (identified by URL's), the ultimate opinion of the
community of a page is reflected in the relative number of links,
taken from across the Web, that ultimately point to that page.
Calculations in this and other cases similar in their basic natures
(although not necessarily involving URL's or stock prices) are
carried out similarly.
[0029] In some embodiments users don't predict specific prices.
Instead they make simpler predictions such as the direction of
movement. In one such embodiment, the user can predict an upward or
downward movement, or make no prediction at all if he thinks the
situation will stay the same or if he doesn't feel he has cause to
predict one way or the other. On this basis, a 0 or 1 is assigned
as the rating, depending on whether the prediction is for a
downward or upward movement, in that order. Then, at the time the
prediction is supposed to have taken place, the system determines
the actual direction of movement. If the stock has gone down, a 0
is assigned as the rating marketplace's rating. If it stayed the
same, 0.5 is assigned. If it went up, 1 is assigned. Then, the
correlation is calculated as described elsewhere between the
individual's rating and the overall rating. In other similar
embodiments, other calculations are used to determine the
accuracy.
[0030] Timeliness is another key value used by the Input Valuation
Unit in many embodiments. It refers to how early a rating is
inputted by the user, since the earlier a rating is inputted, the
less likely it is to be influenced by other people's ratings of the
same item, and therefore the more likely it is to reflect the
user's actual prescience in predicting the ultimate opinion of the
population. Timeliness is not always included in the calculations,
however; for instance, in embodiments where there is no way for
users to "cheat" by only giving their own ratings after they have
seen the ratings of others, timeliness becomes less important or
even irrelevant. An example of this is cases where ratings are not
made public until a certain point in time, when the overall opinion
of the population is calculated and displayed.
[0031] Various embodiments use various techniques for calculating
the timeliness value.
[0032] The most basic calculation simply counts the number of
ratings of the item which are previous to the rating in question;
the "best" timeliness value, then, is 0.
[0033] In some embodiments it is considered to be mathematically
convenient to have a number between 0 and 1. For instance, in many
such embodiments, 1/n, or 1-1/n, where n is the position of the
rating in time (1.sup.st, 2.sup.nd, and so on), is used. In some
others, n/N, where N is the total number of ratings for the item,
is used, which we consider to be a ranking normalized to the unit
interval (and the normalized rank of a randomly chosen user is
uniformly distributed on that interval). One key difference between
these calculations is the fact that the calculations based on 1/n
are scaled such that small differences in the lower values of n
have more import than small differences between large of n. That is
1/1 is quite different from 1/2 (remembering that the range is 0 to
1), but 1/999 is not very different from 1/1000. This is consistent
with the intuitive fact that the user inputting the first rating
for an item has no chance of "freeloading" by copying the rating of
anyone else, while the second user to rate can copy the first
rating; that first rating, if the first user is skilled at rating,
may accurately reflect the eventual average rating of the entire
population; this would enable the second user to appear to be an
accurate rater when in fact he is only an accurate copier.
Similarly, the 3.sup.rd rater can use the average of the first two
ratings, but the value of being able to use 2 prior ratings rather
than 1 is less than the value of being able to use 1 rather than
0.
[0034] The cumulative exponential distribution provides another
technique for calculating a timeliness value that gives more
importance to differences in small n's, and has certain theoretical
advantages as well.
[0035] However, timeliness values based on a uniform distribution
have the advantage of being p-values under the null hypothesis of
randomness, and therefore are amenable to meta-analytical combining
techniques for combining the timeliness with the accuracy and/or
volume of the user's ratings to calculate a statistical
significance, in embodiments where those values, too, can be
considered to be p-values.
[0036] Some values combine the timeliness of the rating under
consideration (for instance, the number of earlier ratings) with
the total number of ratings, due to the fact that when there are
more ratings, the population average is more meaningful. Ranking
(comparing the number of ratings for the current item to the number
of ratings for other items), exponential and other calculations
similar to those already described are used to derive a value
representing the magnitude of the collection of ratings for the
item in question are used in various embodiments; then these two
timeliness-related values are combined by multiplication or other
means.
[0037] When an embodiment involves long-term items such as stocks
which have values that change over time, and for which an
individual rating can be made, at any time or at many times,
regarding the future community rating of the item, some of the
timeliness calculations mentioned above, which assume a first
rating, are not applicable. In such embodiments other mechanisms
are used. One example is a set of embodiments that allow
individuals to make predictive ratings regarding the question of
what the community rating will be at some time in the future, also
specified by the individual. One such embodiment calculates the
difference between the time a rating is made, and the time the
rater expects the community to agree with him (an example being a
prediction that a stock will reach a certain price at a certain
date and time in the future). The average such difference is
calculated for each individual, and than the differences are
ranked. This ranking is made into a unit interval rank and is used
similarly to the other timeliness calculations already
described.
[0038] Thus a user can obtain a better reputation for adding value
to the system by making longer-range stock market predictions,
assuming such predictions tend to be as accurate as those of
someone else who tends to make shorter term predictions.
[0039] The volume value, representing the number of ratings
inputted by a given user, in various embodiments, is also massaged
by means similar to those already described. A rank on the unit
interval, is, in some embodiments, created by comparing the current
user's volume to the volume of ratings generated by other users; in
others the number of ratings generated by the current user is used
to divide one; in others a cumulative exponential distribution is
computed; in others other massaging techniques are used.
[0040] One or more of the values of accuracy, timeliness, and
volume are then combined with each other and/or with other values
to compute an overall valuation of an individual's ratings.
[0041] Some preferred embodiments do this by multiplying values
which have all been normalized to the unit interval and such that
the best value in each case is 1. The product will only be near 1
if all of the factors are near 1, meaning that the user's ratings
tend to be accurate and timely, and that he generates a good number
of ratings. In one preferred embodiment, accuracy is calculated by
the correlation between the user's ratings and the population
average, timeliness is calculated by 1/n, and volume is calculated
by ranking the users according to each user's volume of ratings and
generating a unit interval rank such that, if M is the number of
users, the best rank is M/M (that is, 1) and the worst is 1/M.
These values are multiplied together; the users who add the most
value with their ratings will have a product near 1. In some
variants of these embodiments, weights are applied by taking one or
more of the values to be multiplied to a power. For instance, if it
wants to tune the system to motivate users to generate more
ratings, and embodiment can square the volume value before
multiplying it with the other values.
[0042] Some other embodiments using ranking techniques to generate
unit interval ranks for all values to be considered as p-values
which, under the null hypothesis that all rankings are random, can
be combined via meta-analytical techniques to generate a combined
p-value representing the confidence with which we can reject that
null hypothesis (i.e., accept the conclusion that the combined
p-value is near an extreme because all the values are good). In
some meta-analytical techniques, weights can be applied to the
p-values being combined, analogously to applying weights before
multiplying the values.
[0043] When using the inverse normal meta-analytical method, a
correlation matrix between the three variables to be combined can
be calculated and used in calculating the combined p-value. (This
example should not be construed to exclude other ways of taking
advantage of the correlation matrix.)
[0044] Other embodiments use such techniques as simple arithmetic
averaging and computing a Bayesian expectation.
[0045] The Reward Unit's purpose is to provide incentive for users
to do the filtering and discovery work necessary to unearth the
valuable items, be they stock picks, songs, messages on message
boards, or other kinds of items.
[0046] One set of embodiments involves Reward Units that are based
on providing earlier access to timely information to the people
that add the most value to the system. This is the Early Access
Unit.
[0047] For instance, some embodiments look for cases where several
people are making similar predictions for stock market movements,
where those people have a record for accuracy in past predictions,
and when the agreement of so many people is unlikely to be due to
chance alone. Thus, such a combined prediction may be of value.
Since the earlier an investor knows what is going to happen in the
stock market, the more likely he can make money from that
knowledge, it can be expected that an investor who is also a rater
in a system built on this invention would be motivated to provide a
greater number of earlier, more accurate predictions so that he
will get, in a more timely fashion, the benefit of the system's
searches for unusual agreement among predictions.
[0048] Other embodiments of the Early Access Unit are applicable to
online message boards. Sites such as The Motley Fool's fool.com
allow users to enter ratings for messages in their message
boards.
[0049] Users who subsequently come to the board then have the
benefit of this screening and can avoid messages that are less
highly-rated. This is considered to be a valuable service.
Therefore embodiments which stagger the time elapsed before a user
gets to see a rating provide a way to reward certain
individuals.
[0050] In one such embodiment, the rating summarization to be
displayed on the board are calculated once per minute, in two
different versions. The "elite" version summarizes the ratings
based on all ratings that have been received up to that point in
time. (Summarization takes various forms in different embodiments;
for instance, in some embodiments it simply consists of counting
the number of "thumbs up" ratings received by an item.) Thus, this
elite version is the same as most current presentations. The
"normal" version ignores all ratings submitted within some time
period, for instance, the most recent 15 minutes. Thus, such an
embodiment can present the elite display to the few people who
contribute the most to the system through timely, accurate, and
voluminous ratings, whereas most people would see the normal
version. In other embodiments, many different levels of display are
used rather just two, up to the extreme of having a slightly
different delay period for each user.
[0051] The Early Access Unit is only worthwhile as a motivator in
domains where timeliness is important. A discussion group focusing
on news relevant to the state of the stock market, populated by
serious investors (particularly day-traders), is one such domain. A
discussion group populated by general news professionals, trying to
get the latest scoop before the population in general, would be
another example. Note that in domains such as the one in this
paragraph, a way for a user to improve his timeliness ratings is to
submit his own message, which would be a valuable message, and to
rate it highly. That way he is assured a perfect timeliness value
for that item. (Note that in most such embodiments, the populations
ultimate rating of items submitted by a user is part of the overall
calculation for the value added by the user to the system.)
[0052] Another set of embodiments use a Reward Unit we will refer
to as the Public Reputation Unit. The purpose of this unit is to
reward people who contribute timely, accurate, and/or voluminous
ratings to the system through publicly enhancing their
reputation.
[0053] Various embodiments do this by such means as presenting a
ranked list of the top contributors (akin to the lists of top
scorers seen in many video games) and providing a summary of the
user's contributions when a page dedicated to the user is brought
up.
[0054] Some embodiments provide the opportunity for users with the
highest reputations to be paid for making further contributions in
the form of ratings. The preferred subsets of such embodiments do
this in such a way that it is not possible for a user o motivate
the to "cheat" by doing things like always giving positive ratings
in order to motivate the people who create items to pay him to rate
their new items.
[0055] For example, consider an embodiment in the field of music.
Musicians will want their songs to be gain more attention by
getting a high rating on a listing of new music. But, there is so
much new music emerging every day, just getting enough attention
for users to discover the music and rate it is a major challenge in
itself. Thus, it would behoove musicians to pay raters to rate
their music (that is, as long as the believed that their music was
good enough to be rated highly!) This embodiment of the current
invention presents a list of users, identifying them by their
online ID's and not by their real identities, and listing the
degree to which they have contributed to the system. It does not
show the average of the ratings they have inputted or any other
indication of how high the ratings have been. Instead it presents
the ranking of each user as a contributor to the system, where the
best contributors have the most accurate ratings, the most timely
ratings, and have a high volume of ratings (the calculations for
combining these values in order to rank by the combined value are
described elsewhere in the present document). When a musician is
choosing who to pay, and the musician believes his music will be
rated highly (otherwise he should give up and go home), he will be
motivated to pay the highest-ranked raters, since part of the
rating is accuracy, and this musician believes that an accurate
rating will be a high one. At the same time, the musician is also
rewarding raters who give a high number of timely ratings, even
though the musician may not care about rewarding those things. This
embodiment also provides mechanisms by which raters can post
suggested fees and musicians can enter their credit card numbers in
order to cause money to be transferred to the raters. Various
embodiments in other domains use similar techniques.
[0056] In some embodiments prizes are awarded to the most
contributing user or users; the service displays a notice about a
or prizes that will be awarded to the top user(s) on a particular
date, and this motivates the users to contribute accurate and
timely ratings in order to increase their chances of winning a
prize.
[0057] Another set of embodiments is particularly beneficial for
companies which market products to consumers. A notice is given
that monetary prizes or free products will be given to the best
contributors. The subject of the ratings is messages on a message
board which is devoted to product improvement suggestions. When the
accuracy of ratings is calculated, it is compared not to the
community's combined rating, but to an the value that is assigned
to the item by the sponsoring company. One embodiment assigns a 1
if management decides that a suggestion will be implemented and 0
if it is not, whereas ratings from users have multiple levels
("poor", "below average", "average", "good", "excellent",
translated in the software to 0, 0.25. 0.5. 0.75. and 1, for
example). This mismatch does not impede the calculations, such as
correlation, described elsewhere, although such correlations would
rarely be perfect. Another embodiment allows management to assign a
multi-level numerical rating to each item; management has the
ability to take an item out of consideration when it chooses to, at
which point management cannot change its rating of an item. Up
until then, management can update its opinion based on continued
user input in the discussion group, which would cause accuracy
calculations to be re-executed.
[0058] This methodology for getting suggestions is very attractive
for businesses because it rewards, not only the person who made a
suggestion, but also those who helped to bring it to management's
attention and argue for its importance by rating the suggestion
highly and having an ongoing discussion about it. Note some
embodiments enact this methodology in the message board context
where any message can contain a suggestion tagged a suggestion and
thus enter the process of being judged by management, whereas other
messages, which may be discussions about suggestions rather than
suggestions themselves, are rated and the accuracy of those ratings
is determined as for regular message boards, discussed elsewhere in
this document.
[0059] In preferred embodiments based upon this methodology, a
notice is presented via the user interface to the effect that only
the most early item containing a particular suggestion will be
rated by management as good. That is, if two people present
different messages each containing, in effect, the same suggestion,
only the first one can be rated as good by management. This means
that the second person who likes a particular idea is motivated to
try to improve his ranking by rating the original message highly
rather than by submitting his own item and rating it highly.
[0060] Most such embodiments give extra credit to the originator of
a suggestion, based on the management-supplied rating of the item
which contains the suggestion. For instance, in one such
embodiment, the average management rating for the suggestions made
by each user is calculated, and through ranking against other users
a number between 0 and 1 is generated, then this is multiplied by
the accuracy, timeliness and volume numbers generated as described
elsewhere to determine the overall value of the contributions of
the user. This overall value is then used to determine whether the
user is among the top few who receive a monetary or product reward.
In additional benefit to this methodology is that people who make
suggestions will be motivated to do their own filtering, since they
will be penalized for making poor suggestions.
[0061] In preferred embodiments of the methodology discussed in the
previous few paragraphs, the sponsoring company's record is
displayed. This information includes the amount of monetary award
and/or number and models of products awarded; the time period in
which the awards were granted, and the promises being made about
future awards. This enables consumers to decide if it is worth
there while to contribute their effort to the system.
[0062] It should be noted that various ways of interacting with the
user are implemented in various embodiments. In today's technology,
a particularly convenient set of embodiments is based on the user
interface and networking elements of the World Wide Web. Other
embodiments are based on such technologies as interactive TV,
kiosks, and input and output over the telephone by voice and touch
tones.
[0063] The techniques mentioned in this summary are not intended to
be exhaustive, but rather to point to various examples of ways of
implementing the overall concepts. Other examples are found in the
details section. Other examples are not mentioned, but equally come
within the scope of the invention.
DETAILED DESCRIPTION OF THE INVENTION
Improving Online Community Through Ratings
[0064] There is a lot of "noise" in most Internet discussions in
open forums. Many, people tend to simply state their prejudices
with varying levels of vociferousness. There are also a large
number of very thoughtful, valuable messages. As someone who has
used online facilities both to get information and (in an earlier
lifetime when I had time!) for conversation, I would really like
some way of distinguishing the messages that are likely to be
meaningful ("signal", in the jargon) from the messages that are
likely to be "noise". I would like to increase the signal-to-noise
ratio for my online discussion time. I want to read thoughtful
messages that will really engage me--whether I agree with the
ultimate conclusions of the author or not. Either way, I benefit,
as long as the thoughtfulness is there. (In fact, I am likely to
learn more by reading messages I disagree with, since the author is
less likely to simply be stating things I already know!)
[0065] Several of attempts have been made to accomplish this
through asking users to rate the messages they read, for instance,
on a 5-point scale from "poor" to "excellent". For instance, the
original business model behind Athenium, L.L.C. was to create
NewsVillage, a Web-based overlay on the Usenet, that allowed people
to rate Usenet messages and then read only the ones with the
highest average rating. Other attempts include GroupLens, the
Usenet-rating project begun at the University Minnesota that
ultimately led to the company Net Perceptions, which supplies
collaborative filtering technology for recommending music, books,
etc and is now the recommendation engine at Amazon.com.
[0066] These experiments in message-rating have never really panned
out. They have never "caught on" to allow a service to grow to
critical mass. I think the main reason for this is very simple: it
takes effort to provide meaningful ratings.
[0067] Especially if you read many messages a day, the added effort
involved in rating each one is prohibitive to the point that people
just don't do it because they don't get anything in return for
doing so--except perhaps by being told that they are "helping the
community". If a system forces people to rate a message before
going on to the next one, they will, for the most part, enter
random ratings or enter the same rating every time.
[0068] Anything that adds value to something usually requires work
(there are exceptions, of course, but this is not one of them). In
the "real world", if you do work to add value to a project, you are
paid for that work. If you help to build a house, you are normally
paid in a respected common currency for that effort. Sure, some
people get their friends together to help them build a log cabin in
the woods without any form of compensation other than the pleasure
of being together, but it is unlikely, to say the least, that that
is ever going to be the dominant model for home-building.
[0069] Clearly, knowing what people tend to write the most
rewarding messages, and knowing what messages have been the most
rewarding for the people who have read them so far, would be very
valuable as a way of separating the signal from the noise and
making one's limited time online more valuable. Ratings provides a
means to acquire that information, but the raters have to be
rewarded for their effort.
[0070] One company tried to do this by awarding a tiny fraction of
a frequent-flier mile for each rating a user made for Usenet
articles. I can no longer find that company on the Web, so I assume
that effort didn't go anywhere. I believe the problem with that
effort was that, for enough value to be added in cash dollars or
advertising exposure that frequent-flyer miles could be bartered or
purchased from the airlines, many, many ratings had to be given.
Abstractly but, I believe, realistically, this is because of a
"conversion exchange" in taking items of value in the "economy" of
the Usenet and trying to convert them into items of value in the
"real-world" economy. The conversion rate is extremely steep.
[0071] So we need a reward system that stays within the economy of
the Usenet.
[0072] This economy recognizes several forms of value, for
instance:
a) Respect. People want to be respected for their
contributions.
b) Self-esteem. People want to feel good about their contributions,
which is linked but not identical to winning the respect of
others.
c) Receiving help. People want to be able to ask questions when
they need some help or advice and receive generous and helpful
answers.
d) Interaction. Some people don't care whether they get respect or
not, they just want to interact. Some people actively create
arguments online just for the interaction; hostile interaction can
feel better than none.
e) The pleasure of self-expression.
[0073] All the above are driving values in the economy of online
discussions. Collectively, they are the "currency" of online
discussions. To efficiently pay people back for their efforts, they
should be repaid in that same currency.
[0074] We want to motivate people to provide meaningful ratings. To
provide an efficient reward for their efforts, we need to pay them
back in respect, self-esteem, help, and interaction.
[0075] My idea on how to do this is to calculate the value of each
person's ratings, and to make a very public display of the value of
people's contributions in this sphere.
[0076] People can obviously contribute to the system by providing
meaningful messages
[0077] They are repaid for their efforts by the respect they
receive, the fun of interacting with people's responses, etc. I
have also learned in my years of participating online that people
who are respected members of an online community tend to receive
help, when they need it, much more reliably and quickly than people
with no reputations in those communities.
[0078] For those reasons, people do actively contribute messages to
online communities.
[0079] Now, suppose they received those same rewards for posting
meaningful ratings. Then people would be motivated for creating
meaningful ratings similarly to how they are presently motivated to
create messages--and this implies that meaningful ratings would be
happily created.
[0080] Assume, for now, that we know how to compute the value of
ratings (a technique will be specified using a Bayesian approach
later in this write-up).
[0081] Suppose we post a list of users in order of the value of
their ratings. In our online system, we make this list a very
visible aspect of the system. We use all the marketing know-how we
can muster to communicate the truthful and accurate idea that
ratings are as valuable to the system as messages are,
[0082] because if one has limited time and doesn't know which
messages are the most worth reading, and therefore misses out on
those best messages, they might as well not have been written at
all. Ratings plus messages add up to a highly rewarding experience;
one without the other is less valuable. Meaningful ratings are
therefore to be respected to a comparable (if not identical) degree
to meaningful messages.
[0083] Let's look at how this can be used to reward ratings
contributors in the currencies of online communities.
[0084] a) Respect. The top ratings contributors can win real
respect on the system through the public display of their names
(logon ID's, or whatever identifier makes sense). This is similar
to electronic arcade games, where, although you may not see the top
players actually playing, their initials are stored in the system's
memory, and the high scores and the player's initials are displayed
between games.
b) Self-esteem. Obviously, people who rank above-average in
contributing to the system through their ratings have earned the
right feel good about that.
[0085] c) Receiving help. In a system incorporating this idea, each
person's postings to the discussion boards would include their
level of contribution to the system in a highly visible placement.
So even people who have not contributed at all in the form of
useful messages, but who have provided meaningful value to the
system in the form of ratings, will be the recipient of
appreciation and therefore help from others. This "level of
contribution" does not need to distinguish between the value they
have added by means of ratings vs. the value they have added by
means of highly-regarded messages (which should also be a factor in
computing the level). This reinforces the idea that the value
created by ratings or messages are fungible. As an extreme example
of the importance of being a valued member of an online community,
I once told a friend of mine who needed help with his computer to
log onto a forum on Compuserve in which I happened to be well-known
and to ask his questions there. Over time, I had reached the point
that just about any question I asked received a quick and helpful
response. But my friend was unknown to that community. When he
logged on, he got no response at all to several questions. He came
to the conclusion that online communities didn't work for him, gave
up, and never went back.
d) Interaction. People who are known as contributors to the
community are more likely to receive rewarding interaction with
regard to messages they post.
[0086] e) The pleasure of self-expression. People love to give
their opinions on things. This pleasure has been present in
previous attempts to motivate people to give ratings to messages in
online discussions; now it is supplemented and perhaps overshadowed
by the other forms of currency mentioned here.
[0087] Because the system motivates ratings in the above ways, it
will receive the ratings that will enable people to avoid reading
worthless messages, and to focus their attention on the valuable
ones, thus making their time spent in online discussions
considerably more valuable.
[0088] NOTE 1.
[0089] If the intent is to produce recommendations for the
community as a whole it is important that ratings not be based
primarily on agreement or disagreement with the ideas expressed,
but on their thoughtfulness and/or entertainment value. Of course,
agreement will never disappear as a criterion for ratings, but
appropriate wording in explanations and instructions in the system
can stress the idea that thoughtfulness and entertainment are the
criteria.
[0090] NOTE 2. Not addressed in this write-up is the fact that
ratings can also be used to derive a profile of the rater's
interests, which can in turn be used to compute the expectation of
the value he, as an individual with his own tastes and interests,
would give in rating messages he hasn't read yet. These
expectations can be presented as the priority with which he should
read the various messages in the discussion, which would be
different to the priorities presented to someone with different
interests and tastes.
[0091] NOTE 3. Ratings Input.
[0092] We want to minimize the degree of effort needed to create
ratings. So, the most common case should take no effort at all.
[0093] One solution to this involves a ratings mechanism based on a
pair of "radio buttons". (Radio buttons are user-interface
features, usually a collection of small, circular, clickable
buttons, which allow none of the buttons to be in the "on" state or
one of them to be in the on state, but not more than that.) One
radio button is labeled, "Less thoughtful and entertaining than
average". The other is labeled, "More thoughtful or entertaining
than average". Picking neither button means abstaining.
[0094] NOTE 4. Computing the value of ratings.
[0095] First, we have to decide on what we mean when we say a
rating is valuable to a community.
[0096] We will say a rating is "valuable" to a community if it is
representative of the opinion of that community. That is, if most
people in a community will particularly value a given message, a
representative rating is one that indicates that the message has
above-average value.
[0097] Note that within a particular discussion area, the overall
community can be broken into sub-communities of people who have
similar tastes and interests which might be different from those of
the overall community. All the analysis discussed here can occur at
either the level of the community as a whole or any of
sub-community of like-minded people (see Note 2, above), so this
system can be used to give either customized recommendations of
messages or general recommendations of articles to the whole
community.
[0098] Representativeness is a powerful measure of the value of a
rating simply because the purpose of the ratings is to be able to
predict the community opinion from a small number of votes, so that
people can prioritize the order in which they read messages for
maximum expected benefit. A rating must be representative of a
community's opinion in order to usefully guide that community.
[0099] There are varying degrees of statistical sophistication that
can be used to compute the value of a user's ratings. I will now
describe one very simple approach which is rooted in Bayesian
statistics.
[0100] This takes two steps. First, we calculated the "expected
rating" of each message. We calculate an "expected value" of each
user's ratings.
[0101] Expected rating of messages:
[0102] Some users who won't care about giving ratings. Using the
input mechanism of Note 3, such users will abstain. We will only
consider the actual votes.
[0103] Using the method of Note 3, votes are divided into "Less
thoughtful and entertaining than average" and "More thoughtful and
entertaining than average". We will assign a value of 1 to "more"
votes and a value of 0 to "less" votes. Let p be the ratio of
"more" votes to the total number of votes in the system as a whole,
considering all messages. Using a Bayesian analysis based on the
beta distribution, the expected rating for a message which has
accumulated N actual ratings from users is (pw+m)/(w+N) where m is
the number of people who voted that the messages was more
thoughtful than average and w is a parameter that is adjusted for
best performance as more data comes in from the field (it can start
at 1).
[0104] This means that when nobody has voted on a particular
message, its expected rating is p (for instance, if the system is
equally divided between "less" and "more" votes, the expected
rating for a particular message which has no votes yet is 0.5). But
as votes come in, the expected rating moves in one direction or the
other according to the proportion of "less" to "more" votes. For
instance, if every vote regarding a particular message is a "more",
the expected rating asymptotically approaches 1 (ever-increasing
belief that the next vote will be 1).
Expected Value of a User's Ratings
[0105] We will take a very simple approach here for illustrative
purposes. More accurate approaches can be built, but as the amount
of data grows, their advantages over this very simple approach
becomes less and less.
[0106] We have already noted that a rating is "representative" if
it agrees with the majority of people who have made the effort to
vote on the message in question
[0107] We will assign a value of 1 to a representative vote, and a
value of 0 to a non-representative vote.
[0108] Let p be the proportion of votes that are representative in
the system as a whole. Then, again using a Bayesian analysis based
on the beta distribution, we compute the expected
representativeness of a user's ratings as (pw+r)/(w+N) where N is
the number of votes, r is the number of the user's representative
votes, w is chosen by the system to optimize performance but is
initialized at 1.
[0109] (Note: by optimizing performance, we mean picking a value
that most maximizes the accuracy of the formula in predicting
behavior; this can be easily done using known optimization
techniques once there is a meaningful amount of data.)
[0110] This expected representativeness can be thought of as the
expected "value to the system" of the user's ratings.
[0111] Then, to measure the user's overall contribution to the
system, we can multiply the expected value of his ratings by the
number of his ratings. Users who contribute a large number of
valuable (representative) ratings are rewarded with a high profile
on the contributor's list and impressive contribution rankings on
all their messages.
[0112] This analysis of measuring the representativeness of a
user's ratings has one major limitation, however. It doesn't take
into account the fact that a rating has much more value if it is
the first rating on an item than if it is the 100.sup.th. The first
rating will provide real guidance to all who see the message after
that; the 100.sup.th rating will not change people's actions in a
major way. Also, later raters might choose to simply copy earlier
raters.
[0113] Therefore, we want to weight earlier ratings more than later
ones. The question is, how much more valuable is the 1.sup.st
rating than the second one, and the 2.sup.nd one more than the
3.sup.rd, etc.?
[0114] A reasonable methodology when a small amount of data is
available is to use a cumulative exponential distribution to model
this weight. Taking this approach, we calculate the expected value
of a user's ratings as follows: Let f .function. ( r ) = { 0 ,
rating .times. .times. r .times. .times. is .times. .times. a
.times. .times. hit 1 , rating .times. .times. r .times. .times. is
.times. .times. a .times. .times. miss , ##EQU1## let R be the
collection of all the user's ratings, let c(r) be a natural number
representing the count of ratings other users had assigned to r
before the user in question assigned his, and let
g(r)=e.sup.-.lamda.c(r) where .lamda. is a chosen parameter of the
exponential distribution. A reasonable value for .lamda. would be
chosen such that g(r)=0.5 when c(r)=1; in other words, the value of
a message's first rating is twice as much as the value of its
second rating.
[0115] Then the expected value v of a user's ratings is v = pw + r
.di-elect cons. R .times. g .function. ( r ) .times. f .function. (
r ) w + r .di-elect cons. R .times. g .function. ( r ) .
##EQU2##
[0116] Then the performance of the system can be tuned, once
real-world data is obtained, by using standard computer
optimization techniques such as simulated annealing or genetic
algorithms to find optimal values for w and .lamda..
[0117] The exponential distribution is not necessarily the best
possible basis for calculating g(r). It is not an aim of the
present invention to find the optimal calculation technique for
g(r). Koza's genetic programming technique can be used, another
optimization method may be used, or a genetic algorithm technique
such as the one discussed in U.S. Pat. No. 5,884,282 for evolving a
monotonic function based on a 49-bit chromosome.
A Knowledge Generation System (Kgen)
Introduction
[0118] This write-up describes a technical and business model with
the potential to generate reusable knowledge and organize it in an
accessible manner. The knowledge is thus available for retrieval on
a when-needed basis. The system has some properties that may enable
it to attain exponential growth if used in a public setting,
leading to a high valuation for the company.
[0119] For the purpose of this write-up I will be referring to it
by the brief handle "Kgen".
Ratings and Value-Proportionate Compensation
[0120] Knowledge donation, knowledge valuation, knowledge
organization, and full coverage are all similarly important.
[0121] The Usenet, for example, receives many thousands of units of
knowledge donation per day (in the form of individual articles).
These are of varying value. DejaNews allows the user to retrieve
these articles based on keywords. In other words, DejaNews allows
the user to retrieve relevant knowledge. But it is very inefficient
at enabling useful research to be done, because most of the
knowledge on the Usenet has little or no value, and DejaNews
provides no way to tell which is which.
[0122] It is important, in a truly effective knowledge system, to
have a way to valuate the knowledge items. Knowledge can be
valuated by means of ratings. This important point will be
discussed in much depth later in this write-up.
[0123] Organization is also important. Knowledge is primarily
organized on the Web in two ways: [0124] Hierarchically. It is no
accident that just about every product for which information is
provided on the Web has a hierarchically organized FAQ associated
with it for easy access to information. Hierarchical (or "outline")
organization works well for this purpose, and has evolved to be a
dominant form of knowledge organization on the Web.
[0125] Indexed. Search engines have the ability to automatically
organize knowledge through enabling retrieval based on concepts
(or, more primitively, keywords). However, it is a thesis of this
write-up that such indexed retrieval could be facilitated by means
of human input of appropriate keywords and weights. Indeed,
historically, knowledge collections such as research articles have
always had author-generated keywords associated with them. It is a
thesis of this write-up that the fact that keywords for
Internet-hosted articles tend to be automatically extracted does
not mean that it is preferable to do so, but only that no
motivational system has yet been created to motivate humans to
supply keywords as an additional aid to indexing.
[0126] Organization in each of these ways (or ideally, both
together), makes the knowledge easier to find and thus of more
practical utility.
[0127] Another factor is full coverage of the subject area. This is
important because if the goal is to generate a knowledge base that
people will return to time and time again, they need to
consistently be successful in finding the information they need
there. If they fail to do so too often, they will be frustrated and
spend too much time in fruitless searches, lowering the utility of
the system as a whole.
[0128] Up to now, services on the Internet have focused on easy
knowledge donation (for instance, the Usenet) combined with
automatically indexed retrieval.
[0129] Without a powerful motivational system in place, this is
probably the best that can be expected. Knowledge donation today
occurs in the context of individuals having the pleasure of
communicating with (and sometimes helping) each other. The
knowledge is to a large degree a by-product and thus the
"signal-to-noise" ratio is poor. Automatic indexing, which cannot
distinguish between signal and noise, can retrieve the relevant
knowledge, but not the valuable knowledge.
[0130] There is clearly much room for improvement to this state of
affairs. What is needed is a motivational system that causes people
to contribute low-noise knowledge in such a way that it is stored
in a highly organized way for ease of retrieval. Further, a system
that encourages full coverage would have a further advantage.
Valuation is the Key to Providing a Motivational System
[0131] In order to create a motivational system governed by
something other than social feelings and altruism, people needed to
be compensated for their work.
[0132] Contributing content (knowledge) is work.
[0133] Participating in organizing it is work.
[0134] But, in order for compensation to be used effectively enough
to be useful, it must go where due. Content that is pure noise
cannot be compensated equally to content that is very valuable.
First of all, in that case, there won't be enough compensation to
go around. The system could give every item the same extremely low
compensation as a way of conserving compensation-resources, but
such low compensation won't be enough to motivate the creation of
highly valuable individual items. Moreover, people can get the same
reward for less work by creating worthless content. In fact, an
equal rewards system motivates the lowest level of quality that the
system will award compensation for; anything else would be a waste
of time and energy on the part of the person doing the work.
[0135] (Note that in the above paragraph we are ignoring such
factors as the feeling of doing a job well and peer pressure.
However, these factors are not enough to generally obviate the need
for proportionate compensation in the real world, and there is no
evidence to lead us to believe that the knowledge-generation world
will be significantly different in this respect.)
[0136] So, compensation should be proportional to value.
[0137] Any currency-based economy is an example of this. One
doesn't pay the same amount of money for any object no matter what
it is. A marble costs less than a Ford Escort, and a Ford Escort
costs less than a battleship. In fact, the graduations are very
fine; an "average" marble costs less than a "fancy" marble. All
economies have evolved this way because it enables them to work
more efficiently and effectively. It is a universal principle.
[0138] Kgen needs to embody this same universal principle:
compensation proportionate to value.
[0139] In order to compensate proportionally to value, value must
be measured. Ratings provide a means to do this.
[0140] Ratings are a simple way for value to be determined. From
movie ratings provided by professional reviewers to product ratings
provided by Consumer Reports, ratings are ubiquitous and a simple
way to communicate valuations.
[0141] However, ratings are another form of work. To motivate
non-random, meaningful ratings, value-proportionate compensation
must be provided.
[0142] One attempt to provide compensation for ratings is an
apparently defunct Web site that requested ratings of Usenet
postings in exchange for frequent-flyer miles.
[0143] This is an example of a violation of the "compensation
proportionate to value" principle. The site apparently did not work
well enough to stay around. One possible reason for this is this
compensation principle--the site gave users no motivation to donate
meaningful ratings--it only provided users motivation to randomly
click one of the ratings buttons.
[0144] Without meaningful ratings, we don't know the value of the
donated knowledge items or effort in creating organization. Without
valuations, we can't compensate properly, and can't expect to have
an efficient, effective motivational system.
[0145] (Note: There are other potential sources of valuation, such
as observing the amount of time spent reading a given item. Indeed,
it may turn out that such passive valuation turns out to be a good
adjunct to ratings. However, there is no known example where
passive valuation has been an effective engine for large-scale
valuation of knowledge. Some of its limitations are obvious. For
instance, consider the case of assuming value is proportionate to
the time spent reading an article. The user may go to the bathroom
while reading an article. Or, it may be that one article very
concisely communicates valuable information and so can be read in a
short time, while an alternative article on the same subject is
confusing and rambling and takes much longer to acquire the same
information--in which case the first article would be more
time-efficient and hence have more utility. There may turn out to
be a use for such passive valuation sources, but at present there
is no reason to assume that they can be the primary source of
valuations in an effective knowledge generation system.)
[0146] We need meaningful ratings. Meaningful ratings are work.
Work won't be done without compensation. Compensation must be
proportionate to value.
[0147] So we need a way of measuring the value of ratings
[0148] Rating the ratings themselves is obviously an endless
conundrum. Another solution is needed.
[0149] Such a solution is available. The general principle is that
people who are providing meaningful ratings tend to agree with each
other and to vary their ratings from item to item, whereas people
who don't tend to provide ratings that are random (or,
equivalently, constant). This will be described in more depth later
in this write-up. We will refer to this solution as "expert
convergence".
[0150] By providing a means to estimate the value of ratings, we
can compensate users for providing meaningful ones, thus providing
the necessary foundation for effective knowledge generation.
The Forms of Compensation
[0151] We have established that the heart of the
knowledge-generation system is value-proportionate compensation.
The value of the contributed knowledge and organizational effort is
determined through ratings. The value of the ratings is determined
through expert convergence. So we know how everything is
valued.
[0152] The remaining question is: how do we provide
compensation?
[0153] There is more than one form of compensation available to
us:
Revenue Sharing
[0154] The Mining Company has pioneered the concept of sharing
advertising revenues with the people who are community managers for
various subject areas. There are hundreds of such managers. However
their compensation is typically low (in the hundreds of dollars per
month) as there is not at this time a huge audience to for any one
community that can give a real income. However, revenue sharing is
nevertheless one viable component of a compensation package.
Public Reputation: Respect of Peers
[0155] Public reputation has always been a major motivator in many
areas. People who play video games try to get their initials onto
the lists of the highest-ever scorers that are typically stored in
each machine and displayed at the end of each game.
[0156] Authors and musicians are partly motivated in their efforts
by their public reputations. The same is true for businesspeople
within the community of their businesses.
[0157] Another manifestation of this, and one more to-the-point in
some ways, is the valuable work being done in the open-source
software community on large and successful projects, best
exemplified by Linux. Linux is universally reputed to be more solid
and stable than Windows NT, despite the fact that no money is paid
to Linux developers and NT developers and quite highly paid,
including stock options that have made millionaires of many
Microsoft employees. Open-source developers become known for the
quality of their work within the open-source community. Since there
is no financial compensation, one can reasonably assume that this
reputation is one significant factor in motivating this
high-quality work.
[0158] We can provide visible public reputations to our users.
These will take two forms. One is a top-scorer screen, similar to
those used in video games. The other is a database from which the
score of any user can be seen. Scores will be based upon the value
they add to the system.
[0159] Typically, two scores will be presented: the amount of value
the user has added over time, and the average value per effort
donated (whether a rating, article, organizational effort,
etc.)
[0160] It is expected that these scores will be presented in terms
of percentile ranks within the overall community as such ranks will
be more readily understood than other methods.
[0161] Such examples as Linux seem indicative of the outcome that a
large-scale knowledge base can be largely motivated through
enabling people to earn the respect of their peers through
contributing to the system.
[0162] Public Reputation Earning the Informal Right to Receive
Assistance.
[0163] As a general principle, I have often observed that "newbies"
an online community typically get slower and fewer responses to
questions, if they receive any at all, than people who contribute
often and therefore have reputations for being a valuable member.
People want to help people who are themselves helpers.
[0164] One problem with this way of deciding who to help and who
not to help is that a person may be a major contributor in other
subject areas, and now be in need of help in an unfamiliar one.
Since he hasn't been in that area often, he is unknown to that
community, and therefore thought of as a freeloader--even though he
may be very valuable to the larger community.
[0165] Our scoring system solves that problem by providing overall
percentile rankings for each person which persist as they move from
subject area to subject area. The value a user adds in his area of
expertise will mark him as a valuable contributor.
[0166] The community aspects of Kgen will be described below. The
point being made here is that by earning a reputation for
contributing to the system, one will earn a higher probability of
receiving human assistance when needed in the context of those
communities.
Public Reputation Support for Independent Income.
[0167] All over the word, there are experts that can help with a
given problem you or I may have at a given moment. It may be a
business management question, a programming question, a
recommendation for a home appliance, etc. It could be any of
thousands of things.
[0168] The problem is that we don't know where to go to find
someone who is immediately available to help us and who we can
trust to do a good job. If we had a way of finding someone with the
appropriate knowledge, and they had a reputation that enabled us to
trust that they could help us, in many cases we would be happy to
pay an appropriate fee for their services.
[0169] Kgen's public reputations provide a basis for this trust.
Someone who is highly-ranked within the knowledge area he
contributes to is someone who is likely to be superior at supplying
assistance within that subject area.
[0170] So, by supplying contact information for the contributors,
we help empower them to earn money in exchange for their proven
skill. They are enabled to do so because of the quality of their
work in adding value to the system through one or more of the
possible forms of contribution.
Public Reputation Personal Pride
[0171] Regardless of practical uses for public reputation, people
are also highly motivated by factors of pride and self-esteem. Many
people are highly motivated by finding ways to experience
themselves as good at something. In this case, the additional
factor of being good at something that is contributing to a
community makes scoring highly within Kgen even more meaningful as
a source of pride.
[0172] There are people who spend hours a week as unpaid
co-moderators at AOL, who are given an "official," public status at
AOL--so much so that the question has recently come up as to
whether they should be considered to be actual employees of AOL.
One can guess that a major factor in motivating this work is the
pride of being a leader in the community.
[0173] It is expected that similar factors of pride will motivate
many people to want to earn their stripes as contributors to what
can become the greatest single accumulation of immediately useful
knowledge in the world.
[0174] Again, there can be no doubt that similar motivations can be
found among the contributors to Linux. They also want to contribute
to something large that will be a real force in the world. That
project has succeeded, and it seems quite reasonable to postulate
that this one will too.
Public Reputation: Summary
[0175] It is a thesis of this write-up that these various forms of
public reputation will be a meaningful factor in motivating people
to contribute knowledge, organization, and ratings to the system.
The main benefit of these forms of compensation are that they cost
us nothing.
Other Compensation
[0176] There is no reason that we can't look towards other
compensation mechanisms such as frequent flyer miles as well. These
compensation mechanisms would all be tied to the quality of the
donated work.
Structure of a Public Service
[0177] Some embodiments are primarily structured as an FAQ. FAQ's
have evolved over time to be a dominant form of easily-accessible
knowledge on the Web. As noted above, virtually every manufacturer
with a Web presence has an FAQ describing their product.
[0178] Large FAQ's are hierarchical in nature. They are broken into
various subject areas. To find the answer to a given question, the
user works his way through the hierarchy to find the answer he
wants.
[0179] At the same time, most sites index everything with their
search engines, so the FAQ entries are available through keyword
and concept searches, too. Recent technology such as Apple's
Sherlock and Ask Jeeves enable natural language searches; over time
we should be able to add such technology, either through strategic
partnerships or technology purchases.
Forms of Contribution
[0180] Knowledge items (FAQ entries). Users can write questions and
answers that they think will be useful to the community. These
Q&A's will be considered to be "knowledge items." Users can
also write alternative answers to existing questions. Perhaps they
can additionally write questions without answers (this needs more
thought) in the expectation that somebody will write a question.
[0181] Suggestions. If a user believes an item is in the wrong
place in the hierarchy, he can make a suggestion that it be moved
to another location (each location will be marked through a
numbering system). Easy access will be provided so a user can visit
the suggested location to see whether they think it is suitable.
Similarly, a user can make a suggestion that a Q&A be dropped
due to being redundant. When a suggestion has received enough
positive ratings that it is statistically measured to be highly
likely to be worthwhile, the system will automatically carry it
out. This can involve moving or deleting an item, or perhaps other
acts as the system evolves. [0182] In addition suggestions for
improving an item will inevitably turn up in the discussion forums
associated with that item. [0183] Ratings. Users can rate knowledge
items and suggestions. Ratings of each type of contribution are
considered to be of value. Public Reputation
[0184] Each user's ID will be displayed on each of the knowledge
items he contributes. The ID will be clickable and take the user to
a page displaying the accumulated reputation of the user. His
percentile rank for quality and frequency of contributions will be
displayed.
[0185] A reason not to break these items out is that everyone
realizes that writing knowledge items is a valuable form of
contribution. To date, there is no precedent for the public valuing
of ratings. In order to stress the fact that all forms of
contribution are valuable and worthy of esteem, it may be advisable
to not separate them in the display.
[0186] There will also be means to access a user's scores through
an interface that allows an ID to be typed in and the score
retrieved. Finally, as we add lists of people who want to help
others to each subject area, the scores will be available through
that list.
A Learning-Oriented FAQ
[0187] We will redefine the FAQ, for the purposes of this section,
as Learning Items (LI's) and start to define a different way of
looking at things.
[0188] When students study from a textbook, they frequently have
questions about specific things appearing in the book.
[0189] LI's could work in conjunction with existing course
materials such as a textbook. An LI would explain something that
was perceived to be not as clear as possible. The reference section
on an
[0190] Enter LI screen would contain the page and/or section and/or
paragraph number that was being explicated.
[0191] The idea would be to generate a database that would make it
easier to understand the course material.
[0192] During the Enter LI phase, the team would work together.
Students would each be required to create one LI. However, they are
also in the process of studying, and will have questions. So, on
the team discussion page and/or by chat or other means, they can
ask questions of their teammates when they don't understand
something.
[0193] These questions can then become fodder for someone to write
an LI.
[0194] So, the process of give-and-take during the Author LI phase
is providing a forum where students can help each other understand
the material at the same time as they are working towards creating
their individual LI's, and helping each other with that goal.
[0195] As in the case of FAQ Q&A's, each LI will eventually be
rated by the other team according to how much value it adds: [0196]
Does it add value by explaining something that was harder to
understand in the original text? [0197] Does it do so better than
other LI's on the same question?
[0198] Overtime, the database of LI's for a particular set of
learning materials will become so complete that it would be hard to
create a valuable new LI.
[0199] In that case, as was suggested above, we can randomly delete
some of them. However, in this context, there is no problem
associated with it, because LI's are not general-purpose
just-in-time problem-solving FAQ entries that really should be
available to the company as a whole. They are just for the purpose
of learning, and may only be available to the students taking the
course (or who already took the course). So there is no problem in
removing some of them--there is no outside database to compare to
see what's missing.
[0200] Of course, all LI's for previous rounds can be available to
students or the course through View Results.
[0201] Also, if the instructor chose to use other basic course
materials at another time, there is no reason why all previous LI's
could not be displayed.
[0202] Also, people who already took a course could log on to the
system at any time to refresh their memory about something. So we
would be creating a useful knowledge repository, just as in the FAQ
model, but a user would have to take the course to get access to
it.
[0203] There is still a problem of cooperative cheating, where, for
example, students taking a course at one time copy the LI's and
store them somewhere where they are available to students taking
the course later. A similar problem exists with the SAT's, but
security measures make it very difficult to copy an SAT. These are
issues to discuss.
[0204] One final point is that it can be imagined that through more
elaborate structuring tools, an entire course materials set could
possibly be evolved in this way, which could ultimately replace the
need for outside course materials.
Structure of the Best-Practice Mining System
[0205] Use of the Kgen process for brainstorming and the mining of
best-practices is similar to the learning model, except that study
materials may be unnecessary.
[0206] The ratings processing allows the system to determine which
practices are likely to actually be "best" since rating quality
ultimately depends on the expertise of the rater. Rather than
giving everyone an equal vote about which practices are of value,
including those people who will simply pick the most obviously
correct practices, or who vote randomly, Kgen gives the most votes
to the ratings most likely to be meaningful. Thus Kgen's
discernment is greater than that of a one-vote-per-person
system.
Technical Issues
[0207] Measuring the Quality of Ratings
[0208] I will present a technique for doing this here. I call this
technique "expert convergence." However, I want to stress that it
is very likely that more efficient techniques will be found once we
start serious work in this.
[0209] The calculated quality of a rating is really a
statistically-calculated expectation of the quality. There is no
computational way to know the quality for certain.
[0210] Imagine that we have a knowledge item and a collection of n
ratings for that rating, r1, r2, . . . , m, on a 5-point rating
scale.
[0211] Suppose we make guesses, g, about the real value of the
item, and see how well that guess matches up to the actual ratings
we have in hand.
[0212] First we guess 1/2 point, then 1 point, then 2 points,
etc.
[0213] At each point we calculate the following number: a number
between 0 and 1 representing the probabilistic unlikeliness that
the as many of the ratings would have been as close as they are to
g under the assumption that the ratings were simply randomly chosen
by the users.
[0214] In the patents that have been issued in my name by the
United States Patent Office, I have described detailed algorithms
for accomplishing this. There are other ways, as well.
[0215] One of the points will be calculated to be more unlikely
than any other.
[0216] We will use that point as our estimate of the "real" value
of the item on the 5 point scale.
[0217] Some ratings will be close to this point and some will be
far away. The ones closest to the point are the ones we assume have
the most value, because they are the closest to the estimated real
value of the item.
[0218] A useful measure of the value of the rating would be the
distance from g calculated above (its p-value, in statistical
language; "better" p-values are near 0; p-values are always between
0 and 1).
Measuring the Effectiveness of Raters
[0219] Effective raters tend to consistently give meaningful
ratings.
[0220] Assume, for a given user, we have a collection of p-values
calculated from his ratings of various knowledge items: p1, p2, . .
. , pn.
[0221] Then a meaningful measure, m, of his effectiveness at
consistently contributing meaningful ratings would be given as
(1-((1-p1)(1-p2) . . . (1-pn)) (1/n))), where is the power
operator, for reasons that are beyond the scope of this
write-up.
[0222] Also, under the assumption that his ratings are random, the
likelihood of their being as consistently good as they are by
chance alone can be computed based on the above calculation,
resulting in another p-value. Thus we have a possible statistical
measure of goodness when we want to find the users we are most
confident will produce good ratings.
[0223] Another consideration is the fact that some ineffective
raters will tend to always choose the same rating for each
question. Thus, the variance of ratings provided by a user can be
used as another way of refining our effectiveness measure.
Variances can be converted to p-values, and combined through
multiplication with the p-value determined above; using other
techniques a new "combined" p-value can be calculated that
incorporates both factors.
Measuring the Quality of Knowledge Items
[0224] In order to calculate the likely quality of a rating,
described above, we made guesses about the actual value of the
item, and chose one to be our estimate of the real value of the
item. However, that was done in the context of trying to judge the
worth of a rating, in order to judge the effectiveness of the
rater.
[0225] Now that we have calculated the effectiveness of each rater,
we can do a better job of guessing the actual value of the item. We
can do this be using our knowledge about the rater to weight each
rating.
[0226] For example, we can use (1-m), calculated above, as the
weight of a rating generated by a user with measured effectiveness
m. Alternatively, we can rank the raters according to m, and use
the rank as the weight.
[0227] These weights can be used in calculating a weighted average
of the ratings, which can be calculated arithmetically or
geometrically.
[0228] My collaborative filtering experimentation has shown that
all these approaches will "work", but some will work better than
others. We can deploy or system using the simplest possible
approach, and then try other approaches on the data we collect, and
eventually deploy the calculation that works best for predicting
the ratings of effective raters for given test items.
[0229] Finally it should be noted that iterative techniques are
possible. When we originally calculated an item's best guess of the
real value, in order to measure "expert convergence" for one item,
we did so without incorporating any knowledge of the raters. After
doing that, it would be possible to go back to the beginning of the
calculations are recalculate the expert convergence by taking the
weights into account. That would ultimately change (and improve)
our measure of the effectiveness of each rater; we can repeat until
convergence.
Portfolio Management by Community
[0230] The purpose of this invention is to maximize portfolio
selection by means of creating a open environment for receiving,
rewarding, and integrating trade suggestions from a broad
community.
[0231] A key to the invention is that individuals compete with each
other for being the best performers, which automatically causes the
system to reward them with the best compensation.
[0232] The environment is a network such as the Internet,
client/server systems, etc. Input and output occurs by means of
known techniques of network-based software, such as HTML-based
World-Wide-Web browsers, client/server software on a
packet-switching network, etc.
[0233] One key element of the invention is means for anyone to
register as a "manager" of a portfolio. A portfolio can have any
number of managers. In the preferred embodiment, the registration
process involves obtaining a login ID and password (as is done in
many World Wide Web sites and other online services today) and
specifying the portfolio(s) the user would like to manage.
Portfolios are chosen, in various embodiments, by pull-down menus,
scrolling lists, etc.
[0234] In preferred embodiments, managers can also create
portfolios by inputting the stocks and amounts of the initial
state; this is done by standard input means such as dialog boxes,
HTML pages, etc., as decided upon by the user interface designer in
each particular embodiment. In creating a portfolio, preferred
embodiments give the manager the ability to input his decision
regarding whether he alone will manage it, or whether other people
can manage it as well. In some further embodiments, he can input a
list of users who can serve as managers for the new fund. In must
such embodiments, means are provided to change these decisions and
alter any lists of managers at later times.
[0235] Means are provided for managers to make their trade
suggestions or other input relevant to portfolio management. These
means involve the traditional means for data input at use in most
Web sites. In the case of inputting trade suggestions, a good
example is the user interface of http://www.schwabb.com/.
[0236] The performance of each manager is tracked over time. In
preferred embodiments, this includes tracking the direction and
amount in change in a portfolio's value that would have occurred if
that manager's input, and no other trades, had been acted upon.
These results are measured on a regular basis (in preferred
embodiments, daily, on those days when the manager gave input).
[0237] At regular intervals (in preferred embodiments, daily),
before the market opens on each day, trade recommendations and
other information made by the various managers of a fund are
combined into an overall set of trade recommendations. This is done
by combining the suggestions of the various managers in such a way
as to optimize the gain in portfolio value. Various embodiments
perform this step in various ways.
[0238] In one embodiment, the following steps are carried out for
each portfolio. First, each manager's performance expectation is
computed according to the formula p = t c + n ##EQU3## where p is
the estimated performance, t is the total change in valuation of
the portfolio if only the current manager's input were used, n is
the number of days in which the manager made changes, and c is a
constant.
[0239] For convenience, t is given as the change proportionate to
the starting valuation; that is, if the starting value of the
portfolio was $100,000 and at the end of the period under
consideration, the manager's suggestions would have led to a
valuation of $120,000, then t is +0.2.
[0240] c serves to compensate for the fact that we have different
amounts of data for different users. For instance, say c is 1.
Then, before a manager makes any suggestions the estimate of change
due to his suggestions is 0. If he has made one suggestion, and
that has resulted in an increase of 0.01, then his estimator is
0.005. As he makes more suggestions, c becomes less important, and
the value of the estimator asymptotically approaches his average
change.
[0241] For the next step, we ignore all the managers for whom p is
0 or negative. In this step, we sort all managers in order of p.
Then we calculate weights for each manager as follows. If there are
N such managers, the one with the best performance gets a weight of
2 .times. N N 2 + N , ##EQU4## the next best gets a weight of 2
.times. ( N - 1 ) N 2 + N , ##EQU5## etc. down to a weight of 2 N 2
+ N . ##EQU6## Finally, we take the trade volumes recommended by
each manager for each stock, multiply each such volume by the
recommending manager's weight, and total the weighted recommended
trade volumes for each stock. Depending on the embodiment, the
system outputs the combined recommendation (rounding to the nearest
whole share price for each stock), or automatically causes those
trades to be performed.
[0242] To assign a value for c, preferred embodiments try different
values to see which results in the maximum "yield" in simulations
using the historical database, once there is one. In various
embodiments this can be done by such techniques as genetic
algorithms, simulated annealing, etc. When there is no historical
data or little such data, some embodiments initialize c to 1.
[0243] In another embodiment the following combining method is
used.
[0244] A paper by Singer, et. al. (1998) describes an algorithm for
completely automated portfolio selection. They describe a technique
for using "side information" within their algorithm. Refer to that
document for details. This side information improves the
performance of the algorithm. In the embodiment under discussion,
the advice given by the managers is not concerned with the details
of specific trades, but is rather concerned with providing the side
information. In this embodiment, the side information consists of
characterizing the market for the next market day as "bullish" or
"bearish."
[0245] On each day, prior to market opening (or on the evening
before), each manager has the opportunity to give his guess
regarding which of the two categories the stocks will be in. The
system tracks this information uses it to make a combined guess,
incorporating the guesses of various managers into one.
[0246] A paper by Cesa-Bianchi, et. al. (1997), Warmuth describes
an algorithm for doing this, which they term Algorithm P*. Refer to
that document for details. The input is a series of predictions
from each manager regarding whether the market will be bullish (1)
or bearish (0) on the next day; the output is a real number between
0 and 1. Rounding to the nearer of 1 or 0, the system sets the side
information to bullish or bearish. Details of how to use the
information maybe found in Singer. The output is a set of trades to
make, which is displayed as the output in some embodiments, or is
fed into a module for causing the actual trades to be executed in
another embodiment.
[0247] It should not be construed that this invention is dependent
on any particular combining technology. In fact, a preferred
embodiment provides an application programming interface (API),
using standard industry practices, such that any combining
methodology can be programmed and "plugged in." In once such
embodiment, a C++ base class is provided which can be subclassed to
provide the combining functionality. In another embodiment, a Java
interface is defined that is programmed to provide the
functionality.
[0248] In another a C header file is provided which defines the
signature of a set of functions to provide the functionality. In
preferred API-based embodiments, means are provided to test
different algorithms against historical data to determine their
performance.
[0249] It must not be construed that this invention is dependent
upon any particular combining technique.
[0250] Another key to this invention is compensation of the
managers. In various embodiments, different techniques are used to
perform compensation.
[0251] In one embodiment, a simulations are run with and without
the input of each particular manager, and the amount of growth that
is attributable to his individual input is computed by means of
subtraction. This number is computed for the best manager first,
then the next best manager, etc. A fixed percentage of the
difference is paid to each of them as a fee. This percentage was
determined and input by the manager who originally created the
portfolio in question. Investors are billed on a monthly basis for
these fees.
[0252] In another embodiment, compensation is not directly
financial but instead is time-based. Just as immediate quotes are
considered more valuable than 15-minute delayed quotes, it is more
advantageous to get the combined trade recommendations from the
system sooner rather than later. At the time the market opens for
the day, the best manager receives the recommendation first (in one
embodiment, this is done by online "instant messager"); the next
best manager gets the recommendation second, etc. The intervals are
evenly spaced. The total time allocated to this process is adjusted
so that enough people are motivated to be managers that the
portfolio performs well, but also so that the higher-performing
managers are rewarded for their efforts.
[0253] It should not be construed that this invention is dependent
upon any particular form on compensation.
Notes
[0254] In some embodiments, certain qualifications are required
before a manager can actually make recommendations, such as lack of
a criminal record; in some such embodiments, a manual check is
carried out before newly registered managers are allowed to make
recommendations; in other such embodiments, checking is done
automatically by access to online systems such as credit
database.
[0255] In some embodiments, the system simply receives input from
users and makes trade suggestions. In other embodiments, the system
has built-in software which can manage investor money and order
trades, consistent with other existing systems that currently
perform those functions.
[0256] Means are supplied in most embodiments for funds to be added
to a portfolio at any time for purchase of new stock, as new
investors get involved or as existing investors change the level of
their investments or pull out. Computations regarding the
performance of various managers are adjusted to account for
this.
[0257] In some embodiments, some or all of the managers are not
individual people, but are organizations or even software programs.
The word "manager" in this document should therefore be read
accordingly.
BIBLIOGRAPHY
[0258] Singer, Helmbold, Schapire, and Warmuth (1998). On-Line
Portfolio Selection Using Multiplicative Updates. Mathematical
Finance, 8(4): 325-347. [0259] Cesa-Bianchi, Freund, Haussler,
Helmbold, Schapire, Warmuth (1997). How to Use Expert Advice.
Journal of the Association for Computing Machinery, 44(3): 427-485.
Community-Based Market Movement Prediction
[0260] If one assumes that some people are better at predicting
market and stock price movements (referred to below as "market
predictions" than others, certain opportunities become available.
This invention capitalizes on one of those opportunities.)
[0261] The purpose of this invention is to enable communities to
make more accurate market predictions than individuals can make on
their own.
[0262] The first step is to set up an input means, for instance a
site on the World Wide Web (although any other electronic
communications medium will do just as well), whereby individuals
can enter their market predictions. This is easily within the state
of art of Internet, client/server, and other communications-based
programming disciplines.
[0263] We will use the example of predicting S&P 500 index
movement, although other predictions are possible including
individual stocks. In fact, this invention is not limited to
security markets but is applicable to other markets as well, such
as the price of tulips.
[0264] The system, in preferred embodiments, allows individual
users to input their predictions at any time. At certain points in
time, these individual inputs are examined and group predictions
are made.
[0265] In some embodiments, this prediction is only directional:
users input their guesses regarding the question of whether the
stock will go up or down. In other embodiments, guesses are more
specific as to the magnitude of the movement.
[0266] A key to this invention is measuring the accuracy of each
individual contributor. Different embodiments carry this operation
out in different ways. In one embodiment, in which the user only
inputs directional predictions, a 1 is awarded if the prediction is
correct, and a 0 is awarded if the direction is incorrect; the
average of these numbers is then computed. (In such binary
embodiments, a prediction of no movement is ignored.)
[0267] In another embodiment, a calculation based on Bayesian
statistics is used, We assume a beta distribution, and calculate
the probability of correctness in the N+1th guess, where h is the
number of correct guesses so far, N is the number of guesses so
far, and a.sub.h, and a.sub.i are hyperparameters of the beta
distribution: .alpha. h + h .alpha. h + .alpha. t + N ##EQU7##
[0268] In the preferred embodiment based on the beta distribution,
the beta hyperparameters are chosen so that: .alpha. h .alpha. h +
.alpha. t = the .times. .times. population .times. .times. average
.times. .times. of .times. .times. correct .times. .times. guesses
##EQU8## and a.sub.h is initially set to 1, and refined as
real-world data comes in to optimize the overall accuracy of the
computed probability of correctness. That is, the mean distance
between the Bayesian estimates of user correctness rates based on a
small samples of data, compared to average correctness rates
computed with the complete, large-scale data set should be as small
as possible.
[0269] Another estimator of accuracy would be the statistical
correlation between the real index movements and the guesses. In
this case, the guesses can be bi-valued directional guesses, or
take other forms such as real or floating-point values.
[0270] It should not be construed that this invention is dependent
on any particular estimator of the accuracy of individual
contributors.
[0271] In the next step, the system generates a group prediction
based on the predictions of the individual contributors, combined
with their accuracy levels.
[0272] In one embodiment this is accomplished as follows. Let M be
the number of users who have inputted their guesses on the index
movement. The guesses of users 1 . . . M are then given as g.sub.1
. . . g.sub.M. Also, in preferred embodiments, there is a "weight"
associated with each user, given as w.sub.1 . . . W.sub.M.
[0273] In some embodiments, the weights are simply set to be equal
to the estimators of accuracy. In other embodiments other values
are used. In some such embodiments, we sort the users in order of
the embodiment's accuracy estimator, rank them from 1 to M, and use
the ranks as the weights.
[0274] It should not be construed that this invention is dependent
upon any particular weighting scheme.
[0275] The guesses may be purely directional, multi-leveled or
real-valued.
[0276] In one embodiment, the group estimator is simply the
weighted average of the guesses: w.sub.1g.sub.1+w.sub.2g.sub.2+ . .
. +w.sub.Mg.sub.M.
[0277] In a preferred embodiment, a statistical approach is used
wherein we start with a null hypothesis stating that "Guesses are
random" and calculate a p-value representing the certainty with
which we can reject that hypothesis.
[0278] In some such embodiments, the guesses are massaged so that,
if the null hypothesis were true, they would taken on a uniform
distribution on the unit interval. In one embodiment, this is
accomplished as described in U.S. Pat. No. 5,884,282, authored by
Gary Robinson, hereby incorporated by reference, where we replace
the "ratings" of that invention with discrete levels of predicted
movement; the levels may be labeled, for example "large negative
movement," "medium negative movement," "slight negative movement,"
"no movement," "slight positive movement," etc.; input means are
supplied for users to specify those levels for their guesses. Note
that this introduces a small random element, the effects of which
can be eliminated through integration, as is also described in that
patent. These uniformly distributed (relative to the null
hypothesis) guesses can be converted to z-scores by use of a
standard cumulative normal distribution table (as is also described
in that patent). Let z.sub.1 . . . z.sub.N represent the z-scores
of the guesses (for instance, a negative z-score might indicate a
guess of downward movement and a positive one a guess of upward
movement; the more extreme the z-score the more extreme the guess).
Then z C = w 1 .times. z 1 + w 2 .times. z 2 + + w M .times. z M w
1 2 + w 2 2 + + w M 2 ##EQU9## is also standard normally
distributed [Hedges & Olkin]. Thus, a p-value, p.sub.c, can be
computed relative to the z-score z.sub.c. P-values such that either
p.sub.c or (1-p.sub.c) are very near 0 mean that we can confidently
reject the null hypothesis. This means we can safely assume a
non-random pattern of guessing in one direction or the other. If
there is such a non-random pattern of guessing, the most likely
cause is that there is some degree of consensus among our guessers
that the index will be moving in a particular direction, which if
random chance is not responsible, is probably due to a common
response to available information that the users feel may influence
the market. The closer p.sub.c or (1-p.sub.c) is to 0, the more
confident we can be that some such consensus exists.
[0279] One of the preferred embodiments is similar to the z-score
technique but uses a combining method known as
[0280] It is useful for the system to be able to measure this
confidence. In some embodiments, p.sub.c is displayed unmodified.
Users can then use this number in making their investment
decisions. In preferred embodiments, it is used to generate a
measure that is more meaningful to end-users. In one such
embodiment, data is collected over time regarding the actual
percentage of correct group guesses which have historically
corresponded to various ranges of p.sub.c; for instance, this
percentage is computed for intervals of 0 to 0.0001, 0.0001 to
0.001, 0.001 to 0.01, 0.01 to 0.1, 0.1 to 0.5, etc. (and the same
is done for (1-p.sub.c)). Then, an estimator regarding the
probability that the group guess is correct can be presented to the
user. In other embodiments, techniques such as neural nets or
genetic programming are used to approximate a function that takes
p.sub.c as input and outputs an estimator of historic probability
of correctness.
[0281] Other embodiments use other means of combining guesses. In
particular, Hedges & Olkin, pp. 27-46, give a number of ways of
arriving at p.sub.c from a set of uniformly distributed (under the
null hypothesis) random variables. This invention should not be
construed to be limited to the combining techniques mentioned
above.
[0282] A key aspect of the present invention is its technique for
motivating users to contribute guesses, and, more importantly,
"best guesses", to the system.
[0283] If everyone could have immediate access to group
predictions, there would be little reason for individuals to
contribute their guesses. To deal with this problem, immediate
access is not given to all users. In some embodiments, immediate
access to predictions is restricted to those users who regularly
help the system by providing their own guesses. In other
embodiments, a user who wants the group prediction with respect to
a specific question (such as, "Will the S&P 500 index be up or
down 10 minutes after opening tomorrow?") must input his own guess
with respect to that particular question.
[0284] Preferred embodiments also provide motivation for users to
give their "best guesses" rather than simply save effort by
choosing random guesses. In some such embodiments, the access to
group predictions is staggered in the sense that the "most
valuable" users get access to the group predictions first, and the
"least valuable" users get them last. In most such embodiments,
"most valuable" and "least valuable" are determined by some
combination of the accuracy of a user's guesses and the frequency
of those guesses.
[0285] In one embodiment this is accomplished by ranking the users
for both accuracy and frequency of guesses, and taking the average.
Other embodiments use a weighted average; others use a geometric
mean or weighted geometric mean; other techniques are possible;
this invention must not be construed to be limited to particular
methods for combining these attributes.
[0286] One embodiment does not combine the accuracy and frequency;
instead the timing of when a user receives a group prediction is
determined only by his accuracy; however, he only receives group
predictions for questions he has inputted his own guess for.
[0287] In some embodiments, users can receive multiple group
predictions per question. This is because the system can refine its
prediction relative to a particular question as, over time, more
users input their guesses, and, as the event in question draws
closer, users revise their guesses.
[0288] In some embodiments, users receive one prediction per guess
or revision.
[0289] Various embodiments use different techniques for staggering
users. A preferred embodiment makes group predictions available at
time intervals based on their overall rank on the scale of "most
valuable" to "least valuable" such that there are equal intervals
between the releases of the information. Another embodiment creates
intervals as follows: if the interval between the release of the
group prediction to the most valuable user and the release of the
same prediction to the second most valuable user is x, then the
further interval before the release to the third most valuable is
(1/2)x, the further interval before the release to the fourth most
valuable is (1/4)x, and so on.
[0290] There is some amount of time between the first release and
the last. This can be shorter or longer depending on the needs of
the system. If it is too long, the predictions won't be useful to
the users who are less than the most valuable. If it is too short,
the staggering technique will not produce the necessary motivation
for inputting best guesses. Preferred embodiments allow this time
period to be tuned in real-world use for best results. The quality
of predictions is measured as the staggering is spread out over
varying amounts of time; the normal amount of time is chosen to
maximize the quality (and/or other criteria, according to the
embodiment).
[0291] In preferred embodiments, the total staggering time adjusts
itself according to the nearness of the event being predicted. In
one such embodiment, the staggering time is always a fixed
percentage of the time remaining to the event in question; other
embodiments use other scaling techniques.
[0292] Users can be given access to group prediction by displaying
them on a Web page, emailing them, using an "instant message"
service, etc, In some embodiments, special devices are created for
wireless access to the system.
[0293] In some embodiments, users are only notified about the
strongest group predictions. For instance, in the embodiment
described above using the standard normal distribution, users are
alerted only when p.sub.c or (1-p.sub.c) is less than a certain
value.
[0294] The staggering technique has another purpose. Assume that a
system using this technique turned out to be highly accurate
relative to other predictive means. Large investment houses would
then want to make investments based on those predictions. It is
possible that large enough investments would actually change the
market situation, making the predictions invalid. Some such large
investors may profit as they change conditions, but other
investors, especially smaller investors, whose guesses might be
greatly contributing to the accuracy of the system, would then be
receiving incorrect predictions. This would lessen or eliminate
their motivation for participating, which would make their guesses
unavailable to the system. Since the system is based on the concept
of having a large number of users making guesses, this could be
very harmful to the system's value and functionality.
[0295] Note that "users" of the system can be institutions and even
other computer programs.
REFERENCES
[0296] Hedges, L. & Olkin, I. (1985). Statistical Methods for
Meta-Analysis. San Diego, Calif.: Academic Press, Inc., pp.
370-381. Rating-By-Rating User Valuation
[0297] This appendix discusses aspects of the invention that relate
to certain mathematical calculations
[0298] One problem being addressed is the fact that people can
supply ratings that are essentially random (due to not making the
effort to provide truly meaningful ratings), or which are
consciously destructive or manipulative. For instance, it has been
commented that on Amazon.com, every time a new book comes out, the
first ratings and reviews are from the author's friends, which are
then counteracted with contradictory reviews from his enemies.
[0299] The key to solving this problem is to weight each user's
ratings according to their reliability. For instance, if the
author's friends and enemies are providing ratings simply to
satisfy personal needs to help or hurt the author, it would be
helpful if those ratings carried a lower weight than those of other
users who have a past reputation for responsible, accurate
ratings.
[0300] A problem solved by this invention is to provide a way to
calculate that past reputation.
[0301] This reputation can be thought of as the expected "value to
the system" of the user's ratings. This is bound up with the degree
to which the user's ratings are representative of the real opinions
of the population, particularly the population of clusters which
are more appreciative of the genre into which the particular
artist's work fits.
[0302] (To measure the user's overall contribution to the system,
we can multiply the expected value of his ratings by the number of
his ratings. Users who contribute a large number of valuable
[representative] ratings are, in some embodiments, rewarded with a
high profile such as presence on a list of people who are
especially reliable raters.)
[0303] One can measure the representativeness of a user's ratings
by calculating the correlation between those ratings and the
average ratings of the larger population.
[0304] This analysis of measuring the representativeness of a
user's ratings has s major limitation, however. It doesn't take
into account the fact that a rating has much more value if it is
the first rating on an item than if it is the 100.sup.th. The first
rating will provide real guidance to those who are wondering
whether to download or buy a recording before other ratings have
been entered; the 100.sup.th rating will not change people's
actions in a major way. So early ratings add much more actual value
to the community. Also, later raters might choose to simply copy
earlier raters, so they can mislead any correlation calculations
that way.
[0305] Therefore, we want to weight earlier ratings more than later
ones. The question is, how much more valuable is the 1.sup.st
rating than the second one, and the 2.sup.nd one more than the
3.sup.rd, etc.?
[0306] Let S be the set of all items; let N be the number of all
items; for s .epsilon.S and 0<i.ltoreq.N, s.sub.i is the ith
item. Let u be the user whose rating representativeness we wish to
compute.
[0307] Let g.sub.i,u be the number of ratings received by s.sub.i
previous to u's rating. (i.e., if u gives the first rating for item
s.sub.i, g.sub.i,u is 0.) Let t.sub.i be the total number or
ratings for the ith item.
[0308] Let r.sub.i,u be u's rating of the ith item, normalized to
the unit interval. Let a.sub.i be the average of the ratings for
the ith item other than u's, also normalized to the unit
interval.
[0309] Let .lamda..sub.1 and .lamda..sub.2 be constants.
[0310] Let q.sub.u be the representativeness of u's ratings,
calculated as follows: q u = i = 1 N .times. e - .lamda. 1 .times.
g i , u .function. ( 1 - e - .lamda. 2 .times. t i ) .times. a i -
r i , u i = 1 N .times. e - .lamda. 1 .times. g i , u .function. (
1 - e - .lamda. 2 .times. t i ) . ##EQU10##
[0311] Then q.sub.u is a number on the unit interval which is close
to 1 if the 's ratings have tended to be predictive of those of the
community as a whole, and 0 if not.
[0312] .lamda..sub.1 and .lamda..sub.2 are tuned for performance.
.lamda..sub.1 is a parameter of the cumulative exponential
distribution determining the rate of "drop-off" associated with the
importance of a rating as more ratings for a given item precede 's
rating. .lamda..sub.2 is a parameter of the cumulative exponential
distribution determining the rate at which the drop-off is
associated with the number of total ratings. For instance, if there
are no ratings for an item other than 's, the rating has no
importance in calculating representativeness and is therefore given
weight 0. These parameters can be set manually by intuitive
understanding of the effect they have on the calculation. In some
embodiments they are set by setting up a training situation in
which a number of users rate the items without the means to see
other people's ratings; furthermore, these users are selected and
given financial or other motivation for putting the effort in to
input the most accurate ratings they can generate. These controlled
ratings are averaged. Then standard computer optimization
techniques such as simulated annealing or genetic algorithms are
used to determine values for .lamda..sub.1 and .lamda..sub.2 that
optimize the correlation between these averages and q.sub.u,
q.sub.u is calculated using the entire population of users in usual
viewing mode (such that they could see the ratings of other users).
In preferred embodiments, tuning activities are carried out within
the memberships of individual clusters. That is, the controlled
ratings given by members of a cluster are used to tune the
parameters relative to the general ratings given by other members
of the same cluster. This is carried out for each cluster. If it is
deemed that there aren't enough members of some clusters to
effectively tune the parameters separately for each cluster, then
in such cases the values for .lamda..sub.1 and .lamda..sub.2 are
averaged across all clusters, and clusters without enough members
can use those averaged values. In addition, if a given user has
created ratings in multiple clusters, some embodiments simply use
the average of his representativeness numbers for all clusters as
his single viewable representativeness and some clusters display
separate representativeness numbers depending on the cluster in
which the numbers are being viewed.
[0313] The representativeness of a user is then used for various
purposes in various embodiments. In some embodiments, it is
presented to artists as a reason to pay a particular user to
providing ratings and reviews for new items. In further
embodiments, it is used as a weight for the user's ratings when
calculating overall average ratings for an item. In some
embodiments, listings are provided showing the users' rankings as
trustworthy raters, giving "ego gratification"; in must such
embodiments these numbers are also available when viewing the
user's profile, along with other information presented about the
user.
[0314] It should not be construed that this invention is dependent
upon the particular calculation method for representativeness which
is described above.
[0315] For example, another embodiment uses the following algorithm
for computing the representativeness q.sub.u of user u:
[0316] Calculate the average rating for each item, not counting u's
rating. For each item, rank the population of ratings in order of
their distance from the average rating. In embodiments where
discrete ratings are used (that is, some small number of rating
levels such as "Excellent" to "Poor" rather than a continuous
scale), there will be ties. Simply give each rating a random rank
to eliminate ties. For instance, if the average rating is 3, and
the ratings in order of their distance from the average are, 3, 3,
4, 2, 5, 5, 1, then after randomization one of the 3's, randomly
chosen, will have the top rank, the other will have the next
highest rank, the 4 will have the third highest rank, etc.
[0317] Call the distance from the average, based on these ranks,
the "discrete closeness." Label the ranks such that the closest
rating has rank 0, the next closest 1, etc., up to N-1, where N is
the total number of ratings of the item. Now pick a random number
on the interval (0,1]. Add it to the discrete closeness. Call this
quantity the "real closeness" of user u to the average for the ith
item and label it p.sub.i,u. If user u's ratings are randomly
distributed with respect to the average rating for each item, then
the population of p.sub.i,u's has a uniform distribution on the
unit interval. It can be shown that, due to this, the quantity x u
= - 2 .times. i = 1 N .times. log .function. ( 1 - p i , u )
##EQU11## has chi-square distribution with 2N degrees of freedom. A
chi-square table can then be used to lookup a p-value, p.sub.u',
relative to a given value of x.sub.u. The quantity
p.sub.u=1-p.sub.u' is also a p-value and has a very useful meaning.
It approaches 0 when the distance between u's ratings and the
averages are consistently close to 0, "consistently" being the key
word. Also, as N increases, p.sub.u becomes still closer to 0. It
represents the confidence with which we can reject the "null
hypothesis" that u's ratings do not have an unusual tendency to
agree with the average of the community. So p.sub.u is an excellent
indicator of the confidence we should have that user u consistently
agrees with the ultimate judgement of the community (in most
embodiments, this is the community within a taste cluster).
[0318] Preferred embodiments using the chi-square approach also
include weights relative to how early u was in rating each item and
to take into account the number of ratings for each item. Let
w.sub.i,u=e.sup.-.lamda..sup.1.sup.g.sup.i,u(1-e.sup.-.lamda..sup.2.sup.t-
.sup.i), where g.sub.i,u and t.sub.i are defined as before. Let y u
= i = 1 N .times. p i , j w i , u . .times. Then ##EQU12## p u ' =
Prob .times. { y u .ltoreq. b } = i = 1 N .times. b 1 / w i , u d i
. .times. where ##EQU12.2## d i = ( w i , u - w 1 ) .times. ( w i ,
u - w 2 ) .times. .function. ( w i , u - w i - 1 ) .times. ( w i ,
u - w i + 1 ) .times. .function. ( w i , u - w N ) w i , u N - 1
##EQU12.3## We use p.sub.u=1-p.sub.u' as the measure of
representativeness, with numbers closer to 0 being better, as
before.
[0319] Finally further embodiments provide weights for one or both
of the terms in the expression for w.sub.i,u. Proper weights can be
found using the same procedures as are used for finding
.lamda..sub.1 and .lamda..sub.2; using genetic algorithms and other
optimization techniques, in some embodiments all these weights are
found at the same time.
[0320] In general, in various preferred embodiments of the
invention, various algorithms that allow a representativeness
number to be calculated which includes the predictive nature of the
user's ratings are used, so the invention as a whole has no
dependency on any particular method.
[0321] When displaying the quantities calculated as the
representativeness numbers, preferred embodiments calculate
rankings of the various users with respect to those numbers, or
percentile rankings, or some other simplifying number, since the
representativeness numbers themselves are not intuitively
comprehensible to most users.
[0322] Another useful feature emerges if we take g.sub.i,u to be a
measure of elapsed time in days between the public release of an
item and the time the user rated it (which can be 0 if the review
preceded or coincided with the public release), and
A.sub.2=.infin.. Then the approaches mentioned above for
calculating representativeness can be extended to such situations
as measuring the value of a user in predicting the overall
long-term sales of particular items (or even to predicting stock
market prices and movements and other similar applications).
[0323] For instance, in some embodiments, a correspondence is made
between ratings and ultimate sales volumes. In one such embodiment,
the following algorithm is executed. For each rating level, all
items with that average rating (when rounded) are located which
have been on sale for a year or longer. Then, within each cluster,
average sales volumes for each rating level's items are calculated.
Then this correspondence is used to assign "sales ratings" to each
item based on the total sales of that particular item; the actual
sales are matched to the closest of the rating-associated levels of
average sales, and the corresponding rating is used as the sales
rating. (If there hasn't yet been enough activity in a particular
cluster to conduct this exercise meaningfully, system-wide averages
are used.)
[0324] In this embodiment p.sub.i,u is computed using rankings of
distances from the sales rating rather than from the average
rating. Then .lamda..sub.2 is set to .infin. (in other words, the
(1-e.sup.-.lamda..sup.2.sup.t.sup.i) term is set to 1). Then we
calculate the representativeness, p.sub.u, as before.
[0325] As with the case of calculating representativeness with
respect to general ratings, it should not be construed that this
invention is dependent upon the specific calculations given here
for calculating a user's ratings' representativeness with respect
to sales; other calculations which accept equivalent information,
including the user's ratings, the sales volumes, and time data for
ratings and sales (or, equivalently, elapsed time data), outputting
a representativeness which involves a predictive component, will
also serve the purpose of providing equivalent means for use by the
invention overall.
[0326] For instance, in some embodiments, a rank-based technique is
used for calculating representativeness. In one such embodiment,
time data is used to determine the items that the user rated soon
after their release (or at or before their release) and that have
now been on the market long enough to meaningfully measure sales
volumes. These items are used to perform Spearman rank correlation
between the user's ratings and general ratings or sales volume;
other items are ignored. Other embodiments perform rank correlation
based on this restricted sample and separately perform rank
correlation upon all items rated by the user, and perform a
weighted average on the results.
[0327] Note 1: In some embodiments, it is possible for a user to
change his review and rating of an item over time, since he may
come to feel differently about it with more experience. But for
purposes of calculating, his earlier ratings are stored. In
preferred such iterations, the last rating of an item entered on
the first day that he rated that item is used.
[0328] Note 2: In cases where the cluster has too few ratings or
sales to do meaningful calculations, "virtual" clusters can be
created by combining clusters with similar taste signatures into
one larger clusters for purpose of computing representativeness. In
preferred such embodiments, clusters are successively added to the
original cluster, and the representativeness recalculated as long
as the representativeness number continues to rise with each
iteration. When it declines, this process ends. The maximum
representativeness number obtained in this way is the one assigned
to the user.
[0329] Note 3: In various embodiments the discussed calculations
are conducted at either the "artist level" or "item level". That
is, in some embodiments the artists are rated and calculations done
from those ratings and in others item ratings are used.
A Mechanism for Quickly Identifying High-Quality Items
[0330] The purpose of the invention is to use market mechanisms to
enable new items that are unusually valuable to become known to the
population as quickly as possible.
[0331] The technology domain is computer networks such as the
Internet.
[0332] World Wide Web sites such as HSX (www.hsx.com) currently
allow users to use credits (in the case of HSX, called Hollywood
Dollars) to buy the equivalent of stocks and bonds (in the case of
HSX, called MovieStocks, MusicStocks and StarBonds). These are
traded as in the traditional stock and bond markets. Users compete
to earn the most credits through trading; in some cases, such
credits can be traded for merchandise.
[0333] Because these concepts are fully explained and exemplified
on such Web sites as HSX and Sports Futures (www.sportsdaq.com), no
further explanation while be given here.
[0334] In the rest of this specification, terms such as
"investment," "stock," "shares," "securities," etc. will generally
refer to those terms as used in a system exemplifying the present
invention. However, their basic meanings are taken from analogy to
the stock market and to the other already existing system such as
Sports Futures and HSX.
[0335] The prices of these securities serve as a indicators of the
popularity of the items represented by the security, whether an
album, an actor, or a movie.
[0336] However, these prices are indicators regarding how the item
is viewed by the community as a whole. This is not useful if the
goal is to find an earlier indicator of new items that most people
haven't heard of and that are likely to be popular in the future
when more people become aware of them.
[0337] The first innovation in the present invention is to enable
individual investors to use credits earned through trading to
publicize the items they believe have the potential to be more
popular in the future.
[0338] In preferred embodiments, only credits earned through
investing may be used to publicize the items. In particular,
systems such as HSX give users free credits when they start using
the system in order to invest; however, in preferred embodiments of
the present invention, those free credits cannot be used for
publicizing items.
[0339] The reason for this is to give the option to publicize items
to those individuals who have historically shown that they are able
to determine which items will be more popular in the future. Those
users who have earned a significant number of credits, which are
earned only through prescient investments, have shown themselves
able to make such predictions. It is therefore only fitting that
they should have a greater means to publicize items than users who
have not demonstrated such ability.
[0340] In one embodiment, publicity is achieved by means of a list
of items sorted by the number of credits being spent to publicize
them. (mp3.com currently has a feature where fans of artists can
bid money to list their items on a similar list.) Investors are
thus, in effect, "bidding" for the top spots. Bids from more than
one investor for the same item are added together to produce a
great total bid for the item. In various embodiments, this list can
be of fixed length, with only a predefined number of spots
available (in which case only bids that result in winning a spot
would actually be charged to the user's credits), or of unlimited
internal length, with each user choosing how many spots they want
to see. That is, a user might set his system profile to show only
the top ten spots on each list. Other embodiments allow users to
specify a minimum number of credits such that only items with at
least that many credits bid on them will show up on the list. The
list is available for all users to view, in a prominent and
easily-reached spot on the service.
[0341] In another embodiment, publicity is achieved by such means
as emailing information about the current top item (the one for
which the most credits have been bid to publicize the item) to
users. In some such embodiments this occurs on a daily basis; in
others, the emailings are more frequent; in other embodiments,
"instant messager" or "pager" software is used.
[0342] In another embodiment, publicity is achieved by enabling
investors to bid to have the regular listings for items to be
highlighted in some way, for instance, by showing them in a
different color. In some such embodiments, only a fixed percentage
of items can be highlighted in this way.
[0343] In embodiments where a publicity list is included, the
preferred such embodiments include means to make the list available
outside of the system itself. These means are accomplished by such
means as XML, XML-RPC (XML Remote Procedure Call), SOAP (Simple
Object Access Protocol), JRI (Java Remote Invocation), CORBA, and
other techniques which allow separate software systems to
communicate information.
[0344] For instance, in cases where the items are musical
recordings, the publicity list might be made available to various
Web sites which focus on MP3 and other music-related issues. This
would enable them to help their users find new high-quality
recordings more efficiently.
[0345] In most embodiments, spots are purchased on the publicity
list for a fixed amount of time, for instance, a day or a week.
[0346] In preferred embodiments, means are provided to filter the
publicity information (whether it is represented through a
publicity list or through other means such as color) according to
the tastes and interests of the viewer.
[0347] For instance, in one embodiment, there are separate
publicity lists for each category of item. For example, in a
music-related system, each musical style has its own publicity
list. In some such embodiments, the categories aren't rigidly
associated with the items themselves, but rather, users are given
the means to specify what category they want their credits to go
toward listing an item in.
[0348] In preferred embodiments which include categories, users can
use some of their credits to publicize an item in one category, and
use further credits to publicize the same item in one or more other
categories. This is useful when there is no single clear category
for an item; for instance some recordings could be categorized
equally well as rock or jazz.
[0349] In some embodiments, other means than categories are used to
filter the publicity information for different users. For instance,
the techniques described in U.S. Pat. No. 5,884,282, incorporated
herein by reference, assign an appropriateness to each item, which
is between 0 and 1, and which is relative to the tastes of the
individual user viewing the output of the system. In that invention
these numbers are used to order recommendations of items such as
movies from the ones the user is expected like least to the ones he
is expected to like most.
[0350] These appropriateness numbers, which are computed relative
to the individual viewer's tastes, can be multiplied by the number
of publicity credits achieved by each item, resulting an recomputed
publicity credits which are skewed to account for the individual
viewer's tastes. Since in the present invention we expect the
publicity credits to have a strong correlation to the quality of an
item, the term "goodness" is In various embodiments, various other
means of combining these two numbers--the publicity credits and the
individualized goodness--can be used, such as computing an
average.
[0351] In some embodiments, the following reasoning is taken into
account. To get a high position on the publicity list, the placer
should not use points (points are discussed here for convenience
but the reasoning regarding using money is similar), because once
he gets a lot of points, the number of points has more to do with
his past success and his risk comfort zones than it does with his
actual confidence in the particular item. Instead, he should risk
his reputation when he puts an item high on the list. For instance,
in a preferred embodiment, his position on the list is due to the
confidence level he expresses for that item. Standard means are
provided for him to specify this confidence level; on a Web-based
interface, it may take the form of a pull-down list describing
various confidence levels from high to low, or a text input area
where a number may be typed. Then the weighted average of how
people rate items he put on the, where the weight is based on his
confidence level. (These ratings may be collected independently
from the list itself, for instance, in another portion of a Web
site where people rate items they have heard.) This weighted
average becomes his reputation. If he's highly confident, he can
get a high list placement, but if other people disagree, it will
hurt his reputation. For instance, there can be 5 confidence
levels, and if he gives it a 5, he will probably get the top spot
on the publicity list, but the weight on ratings given for that
item may be 5 times normal. In preferred embodiments, the
reputation is displayed next to each entry on the publicity list.
In one such embodiment reputation consists of the weighted average
of ratings described earlier in this paragraph, and is displayed
next to the entry in the publicity list. In other embodiments, the
average ratings are translated into a reliability measure; in one
simple form, this measure is displayed as "mostly reliable" or
"mostly unreliable", depending on whether the weighted average of
ratings is less than average or greater than average.
[0352] In other embodiments, input means are provided, such as a
button, in the publicity list next to each entry, where people can
rate the appropriateness of that entry.
[0353] In some other embodiments, the maximum sales ranking
(computed using similar means to old "top-40 lists", where the
sales over a particular period of time such as a week are ranked)
attained by an item over some (longer) period of time is computed
and used to judge whether the item should have been recommended and
thus to construct a reputation for the person who placed the item
in the publicity list. For instance, for 6 months, week-by-week
sales are measured, and the week of greatest sales is considered. A
recommended item may be required to reach some particular level,
for instance, to achieve sales in the top half of weekly sales
ranks when the best week of every item is considered. If it does
not achieve the required level, the lister's reputation goes down,
otherwise it goes up. For example, in some embodiments, 1 times the
weight is used if the required level is achieved, and -1 times the
weight is used if the required level is not achieved. Then these
numbers are summed and divided by the sum of the weights to compute
the weighted average. This number is then translated into some
easy-to-interpret form, such as "mostly reliable" or "mostly
unreliable", as described earlier, or some other form.
[0354] Another way of looking at it, however, is this: the top spot
on the list gets the most looks and therefore the most ratings, and
therefore can affect the user's reputation the most . . . so no
weight needs to be used; we simply average all the ratings given to
any of the items the user has placed in the list; we average them
all together to come up with a reputation.
[0355] In some embodiments, an item may only have one listing on
the publicity list, even though more than one person wants it
listed. This leads to the question of how to combine the
reputation-related risk undertaken by multiple listers into one
listing position, and how to calculate the changes to the
reputation of each lister.
[0356] One solution, used in preferred embodiments, is as follows.
Users are enabled to assign a special kind of point--we'll call
them "repupoints"--to each item they want to list. Users can assign
some fixed range of repupoints to any item, for instance, between 1
and 5. Users can assign repupoints to as many items as they want,
without limitation. These are summed for each item; items are
placed in order of the total number of repupoints for each item.
The number of repupoints assigned by each user is that user's
weight. In such embodiments, the combined reputations of the
listers is displayed with the entry in the publicity list, for
instance, by averaging the reputations.
[0357] If their reputation drops below a particular level, they can
no longer assign the maximum number of repupoints. Below another
lower level, they can not assign more than a particular (lower)
number of repupoints, etc.; in some embodiments this is a
continuous scale, while in others there are fixed steps. This
technique limits the damage to the system that can be done by
people who don't care about their reputations, while still enabling
people whose reputations are poor to improve them by placing
listings in lower slots. This way they can improve their
reputations without risking the time of as many other users, who
don't necessarily look below the top tier of listings.
[0358] In preferred embodiments, the lister's reputation
calculation system is not based on how good the item is, but
instead is based on how much the popularity of an item improves in
the period after the listing is placed. In such embodiments there
is a "popularity improvement measurement mechanism." This invention
is not limited to any particular way to calculate this, but an
example will be given for convenience. Listings generally only have
a limited life, for instance, one week. In one simple popularity
improvement measurement system where listings last 1 week, it is
required that the average number of downloads of the item per day
in the period beginning 3 days before the listing began and ending
on the 4.sup.th day of the listing is less then the average number
of downloads per day in the period beginning on the 4.sup.th day of
listing and ending on the 3.sup.rd day after the listing. (Other
embodiments may use other time periods, according to the particular
application and judgement of the designers.) In this way, the
listing was shown to have value, since downloads increased
correspondingly to the presence of the listing. (In order to
increase the accuracy of this mechanism, it is helpful to have
information such as reviews, "chat" and discussion-group
discussions, and ratings of the item easily accessible from the
publicity list entry. For instance, in some embodiments the average
of user ratings for the item is presented in the publicity list. If
the ratings aren't good, the presence of the item in the list can
actually have a deleterious effect on the number of downloads,
since more people may see the negative ratings.) Following one
mechanism described above, if the popularity increases, we give
1*the weight, and if it decreases, we give -1*the weight; we add
these numbers for each week and item the user lists, and divide by
the sum of the weights, giving the reputation.
[0359] Some embodiments take a slightly different approach at the
cost of some additional complexity. The lister inputs a percentage
of people who will rate the item "worth having been downloaded" (if
the rating is collected later) or who will simply download the item
(for embodiments where passive rating collection is desired).
Placement is based on this percentage; the higher the percentage,
the higher in the list. If multiple people want to list an item,
the percentage can be averaged between the listers; the estimates
supplied by each lister can be weighted according to the reputation
of each lister. If the listers are too conservative in their
estimates, the entry will have a low placement on the list and
maximum benefit will not be attained. If they are too high in their
estimates, the system then punishes them by lowering their
reputations. In preferred embodiments the system punishes those who
overshoot their estimates by a great amount more than those who
only slightly undershoot it. In one of the simplest embodiments,
there are no weights, and the system assigns I if the estimate is
equal to or less than the reality, and -1 if it is less than the
reality. In another embodiment, the system converts all percentages
to the equivalent number between 0 and 1 (0.75 for 75%, for
example), and calculates the reality minus the lister's estimate.
If for instance, the lister estimates 75% and the reality is 90%,
the "score" will be 0.15. If he estimates 90% and the reality is
10%, the score will be -0.8. These scores are averaged for each
user to determine the reputation, which, as described elsewhere, is
usually subsequently converted into an easier-to-understand form
for display purposes.
[0360] It must be noted that publicity lists involving user
reputation calculations as described here can be used in many
contexts other than the ones described in this invention
disclosure, and therefore may be considered to constitute an
independent invention.
[0361] Another innovation of the invention lies in providing means
for manufacturers to publicize their new products while
simultaneous rewarding the participants who add the most value to
the service.
[0362] In one embodiment this is accomplished by enabling users to
buy new items with credits instead of using money. The users who
have the most credits to spend are the ones who have added the most
value by helping to publicize items that eventually become popular,
and who the item producers would therefore assume would be most
likely to be able do the same for their new items. Thus, both
parties benefit.
[0363] In further embodiments, only credits earned through
investing can be used to buy items. The software keeps track of how
many credits each user has earned through investing and uses that
tabulation to restrict purchases to credits earned in that way.
This further ensures that only users who are skilled at discerning
and publicizing high-quality new items will purchase the items.
[0364] In some embodiments, purchases are not made using credits,
but rather, item producers simply give new items to those who have
earned the most credits. The system makes a display available which
enables item producers to see who has earned the most credits (for
instance, in some such embodiments the listing of users is sorted
by total amount of credits earned through investing); in some such
embodiments the display includes such information as email address
or name and address where items, such as musical CD's can be
mailed. In preferred further embodiments, only qualified item
suppliers are allowed to access such listings for privacy reasons;
this is accomplished using such means as supplier logon using
digital certificates. In some further embodiments, sending items to
users is accomplished automatically through interaction with the
item supplier's computers by XML-RPC, or other means.
[0365] In some embodiments, similar means to those mentioned above
are used to discern the most valuable users, and those users are
paid money in return for carefully considering particular
items.
[0366] By motivating the most valuable users to consider particular
items, it can be expected that those users will choose to invest
credits in the more valuable such items, thus making a credit
profit if the item eventually becomes more popular; they can also
be expected to use some of their credits to publicize such
items.
[0367] In some embodiments, items consist of information generated
by the users themselves. For example, in one embodiment, the
invention is used as a brainstorming tool. Users have a problem to
solve, and generate ideas that might be used to solve the problem
with the goal of deciding on a few "best" solutions. One problem
with traditional brainstorming methods is that they don't scale
well to a large number of people--which is unfortunate because the
more people there are, and the more different backgrounds and
abilities they have, the higher the likelihood that someone will
generate an optimal solution, particularly through active
interaction with the input of other people. A problem with
traditional methods as they scale to include a large number of
people is that it is hard for particular ideas, even if
extraordinarily good, to emerge from the noise. The person whose
ideas are most likely to emerge is not necessarily the person with
the best ideas, but rather the person with the most social skill
and convincing manner. The present invention serves to create a
very dynamic market-based solution, where the velocity at which a
good idea can spread is greatly accelerated by the market process,
giving more people the chance to interact with that idea more
quickly, thus generating further enhancing ideas more quickly,
greatly increasing the rate at which the entire process occurs.
[0368] In some embodiments, such as the brainstorming one just
mentioned, an input form is provided, for instance, by means of an
HTML form, for the user to input an information item. It may be an
idea, it may be an article, or even a humorous piece.
[0369] The item is initially assumed to be worth a certain number
of credits, for instance in some embodiments every new item is
assumed to be worth 1,000 credits.
[0370] In some embodiments, the initial valuation is based on the
history of the item creator. For instance, suppose the average
current item valuation is X, the average valuation for items
created by the creator of the item in question is Y, the number of
items created by that user is N, and W is a weighting factor. Then
one embodiment uses the formula (WX+NY)/W to calculate the initial
value for new items. W may be chosen using various means in various
embodiments; in one embodiment, standard computer optimization
methods are used to determine the value of W such that, using
historical data, at the end of 24 hours of a new item's
availability, the output of the formula most closely equals the
valuation as determined by the marketplace. In other embodiments,
other formulas are used. For instance, in one embodiment, a
Bayesian calculation using a multinomial distribution based on the
Direchlet distribution is used. This involves breaking up the
possible valuations into discrete ranges. Any competent
practitioner of Bayesian statistics has the know-how to generate an
appropriate algorithm for this.
[0371] In some embodiments, an initial valuation is not considered
to be required. Instead, the originator of an item chooses to make
a certain number of shares of the item available on the
marketplace, which are auctioned. For instance, the creator of a
new musical recording might make 10% of its shares available, which
can only be purchased in fixed lots of 1% each. These lots are
auctioned using standard auction mechanisms or market-making
mechanisms. The creator receives the credits paid for the shares.
This mechanism provides an initial, tentative valuation of the
item: after this first round of stock sales, the item in this case
can be considered to be worth 10 times whatever the shares sold
for. However, the people who bought those shares will now be
interested in helping to publicize the item, since they believe the
item has worth and that when it is better known, shares will be
sellable at a higher price in the future.
[0372] In some embodiments people desiring to sell shares can
choose whether to put them up for auction or to sell them at a
fixed price.
[0373] Market-making mechanisms for facilitating the trading of
stock are already in use at HSX, Sports Futures, and elsewhere, and
will therefore not be described in detail here.
[0374] However, it should be noted that use of a valuation device
which exists outside of the marketplace itself can facilitate
market-making and provide a standard by which shares of a given
item can be translated into the external world (for instance, into
cash).
[0375] In some embodiments which involve the online music industry,
for example, the relative download rates of various items are used
to provide the valuations of those items. The external valuation of
an item is always proportional to its current download rate. This
external valuation is used by the system to enable the system to
buy and sell shares which are offered for sale or purchase
commensurate with that valuation. This helps the system play the
role of market maker.
[0376] In other embodiments, other mechanisms are used to create
the external valuation.
[0377] In one such embodiment, a list is kept which represents the
current population-wide popularity of an item. For instance, in a
brainstorming-related embodiment, this list shows how important
each item is thought to be by the population. Each user is given
input means to enter a current importance number for each item,
between 0 and 100. If a user does not enter a number, his entry is
assumed to be 0 (it has no importance to that user). These numbers
are averaged to compute the popularity of each item; they are
listed in order of popularity, and this popularity measure is also
used to form the external valuation.
[0378] In another embodiment, the external valuation is
proportional to the number of hyperlinks going to an item. For
instance, this is an appropriate technique when the items are Web
pages or sites.
[0379] In another embodiment, the external valuation is
proportional to the number of items on user hard disks. For
instance, Napster is a process that runs on the computer of each
user. It can be modified to count the number of MP3 files on the
hard disk of each user and send that information (preferably
anonymously) to a central server where the results can be tabulated
(preferably, only one occurrence of each MP3 file on a given user's
machine is counted). Thus, if the present invention were integrated
with Napster, for instance, in order to quickly identify the best
new MP3's, this external measure would be practical. Integration
with Gnutella could proceed similarly, although in that case the
range of types of items is much greater.
[0380] In another set of embodiments, limitations are placed on how
much input users can give to external valuations.
[0381] In some such embodiments, users are given an ability to
express their opinions regarding external valuations proportional
to the amount of credits they have earned through their
investments. For instance, one such embodiment allows users to
allocate various percentages of their earned credits toward
external valuations.
[0382] For example, let us assume a minimal system for ease of
explanation. Suppose two users have earned credits in the system.
User A has earned 100 credits, and user B has earned 200 credits.
Suppose there are three items for which external valuations are
desired. If A allocates 25% of his credits to item 1, 25% of his
credits to item 2, and 50% of his credits to item 3; and B
allocates 50% of his credits to item 1, 25% to item 2, and 25% to
item 3, then item 1 will have 25+100=125 credits, item 2 will have
25+50=75 credits, and item 3 will have 50+50=100 credits. These
credits, then, comprise the external valuation of the items. Note
that using percentages for these allocations has the advantage that
the amounts can be recomputed dynamically as the number of credits
earned by various individuals rises and falls.
[0383] An alternative method is to give each individual a certain
number of points to allocate for external valuations, which does
not change over time. However, this method does not give more
weight to people who have demonstrated their prescience. In some
compromise embodiments, each user has a fixed number of points
given to him, which are added to proportionate to his earned
credits. Other variants also fall within the scope of the
invention.
[0384] Note that these allocations, in preferred embodiments, are
not time-limited. A user can continue to allocate a percentage to a
particular item over time. This helps the external valuation be
more constant and resistant to short-term trends.
[0385] Other embodiments use other means for dividing credits or
points for external valuation purposes. For instance, in one such
embodiment, the user simply lists his favorite items in order of
how high he things they should be valued. (Various input means can
be used to accomplish this, for instance, "dragging and dropping"
items to place them higher or lower in a visual display where a
higher position means an item is to be more highly valued; this
could be accomplished, for instance, by means of a Java applet.)
Then, an algorithm runs that allocates 1 unit to the lowest item on
the list, 2 to the next highest, and n to the highest (assuming
there are n items in total). Then the total number of allocated
units is computed. Then for each item i where I=1 . . . n, where ui
represents the number of units allocated to the ith item, U
represents the total number of units allocated by the algorithm for
the current user, and C represents the current user's current
number of credits, and ci represents the number of credits
allocated by the current user toward the external valuation of item
I, ci=(C*ui)/U.
[0386] Users who have earned more are given more rights to buy
space on the publicity lists for the items they favor, but, in some
preferred embodiments, rather than giving such users the right to
bid proportional to their earned points on individual items, each
person has a maximum of one "vote" per item (or another relatively
small number of votes); but the more points a person has earned,
the more items he can vote on. For instance, it might cost a
certain number of points to buy a vote, but only one vote can be
bought for a particular item. In some embodiments, these votes are
purchased with real money, but the right to purchase them can only
be acquired by earning points through the investment process, which
may be based on a non-money system.
[0387] As elsewhere in this document, the examples given are
intended to be illustrative and not limiting.
[0388] In some other embodiments, usually one involving audio or
video items which are available for the public to hear or view, and
in which the items are played consecutively without user choice
about the order the items will be played in, means is be provided
for users to "skip over" the rest of an item after they have heard
part of it. The valuations of the various items are inversely
proportional to the number of such skips the item receives. An
example would be a Web site that plays songs one after the other,
as a radio station would, but which enables users to skip over ones
they don't like by clicking an HTML button on the site.
[0389] In some embodiments, there is a minimum allowable initial
value for new items.
[0390] When used as a brainstorming tool, some embodiments only
provide an external valuation at two points: first, when an item is
created at which time it is some minimum number (or a number that
is a factor of the total number of items, for instance, one divided
by this number), and when a trading session is over the idea is
finally accepted or rejected as an actionable idea, at which time
the valuation the sum of the initial values of all items divided by
the number of accepted items; the rejected items each have a value
of 0 credits. (This differs from the mechanism used by Sports
Futures, where each "contract" is worth $0 or $100 at the end of
the season in that it accounts for the fact that new items may be
created at any time.)
[0391] In some embodiments, a mechanism is provided through which
the user must spend credits to create a new item. This ensures that
users are careful to create items which they believe have quality.
In one such embodiment, a user who creates an item is required to
buy all the shares of that item at the item's minimum allowable
valuation. He may then make any number of those shares that he
chooses available for sale. In some related embodiments, the item
creator is given a special price he can buy the shares at which is
not necessarily associated with the item's minimum allowable
valuation. In some related embodiments, a fixed proportion of the
shares, for instance, 1/2 of the shares, are always made available
for sale and the item's creator keeps the rest.
[0392] In preferred embodiments, each user gets a regular
"allowance" of credits, for instance, on a weekly basis, that he
can use to buy shares and/or to publicize items. Additionally, in
preferred embodiments, each user gets a certain number of credits
when he first creates an account with the system. Some embodiments
display prohibitions about the same person creating more than one
account (thereby collecting the initial allocation of credits more
than once), including stating the right to terminate any duplicate
account. In some embodiments, digital certificates from
3.sup.rd-party companies, which try to guarantee uniqueness, are
required as part of registration. In some embodiments, cookies are
used to ensure that there is only one user per computer.
[0393] In some embodiments, there are different categories of
credits. In some such embodiments, credits received in the initial
allocation, and/or credits received from a regular allowance, are
not usable for publicizing items; in such embodiments, credits
earned through profitable trading can be used for that purpose.
When a trade is made, the difference in credits between the
purchase price and the sale price is allocated to the fund which
can be used for publicizing items. This difference can be negative
and will thus lower the credits available in that fund.
[0394] In some embodiments, a marketplace is created for a fixed
amount of time, for instance, when the invention is used for
brainstorming to solve a problem which needs to be solved by a
particular time.
[0395] In some embodiments, mechanisms are provided by means of
which users can receive cash payments in exchange for their
credits. For instance, a check can be created by the computer
system and subsequently mailed; in others, software and network
systems can be used to cause the sum to be placed in the users
credit card account. In some embodiments, the translation of
credits to cash occurs at the user's discretion; in some
embodiments with time limits, all credits are converted to cash at
the end of the marketplace's lifetime. In some embodiments, only
certain types of credits are exchangeable for cash, for instance,
credits earned through trading (as described elsewhere in this
application).
[0396] In some embodiments, the invention is framed as a
competitive game in order to further motivate the users. For
instance, a scoreboard is available where various players' current
scores are listed; credits are considered to be points. In some
such embodiments, only credits earned from trading are translatable
into points.
[0397] In some embodiments, the creator of a session is given means
to specify certain requirements for participation that are related
to the users' past performance in other sessions. For instance, if
a new brainstorming session is created to solve a particular
problem, some embodiments might enable registration for the session
to those who have previously earned a certain number of credits
through trading, or users may be required to buy their way in with
credits, or other restrictions may be applied.
[0398] In most embodiments, discussion software user interfaces are
attached to the items. That is, when viewing an item, the user can
see comments written by other users about the item, or add comments
to those already written; alternatively, the comments are a click
away from the display of the item.
[0399] In some embodiments, shares pay dividends, as shares in the
regular stock market do.
[0400] In various further embodiments, shares and/or their
associated dividends are purchased and paid in real money.
[0401] As an example, in one embodiment an item is a downloadable
song which is paid for by the downloader, as can be done with
Liquid Audio's software. Each song is divided into 10000 shares.
Revenues derived from sales of the song, after overhead costs such
as a fee to Liquid Audio, would be paid to shareholders as
dividends. The entity which owns the actual song may choose to keep
a certain number of shares for himself, and to make the rest
available to the market. He may set an initial price for the
shares. Users of this embodiment may then purchase shares. Since
the real value of the shares would be based on expectation of
future dividends, and this would be based on sales volumes, market
forces would be expected cause the share price to generally be
proportional to current sales. In this embodiment, a publicity list
is also provided whereby interested parties can bid real money for
high placement. Shareholders in a song by a new and unknown artist,
who bought many shares cheaply, and who believe in the future of
the artist, would be very motivated to raise sales by publicizing
him in this manner. The sponsor of the trading system itself
profits by taking a portion of all revenues, as well as by keeping
the fees paid for the publicity list.
[0402] Other, related, embodiments are different in various aspects
from the one described above. Various embodiments focus on other
domains than music. In some embodiments the number of shares is
determinable by owner of the item. In some the number kept by the
creator of the item is fixed, for instance, 50%. In some no
publicity means is provided; in some, other means than the
publicity list are provided. In various embodiment, various
procedures are used for the system owner to make a profit,
including keeping publicity bids, keeping a transaction fee for
trades, etc. It should not be construed that this invention is
limited to such details of implementation.
[0403] In some embodiments, an information display is publicly
available which states the amounts of current dividends. In typical
examples of such embodiments, dividends are calculated on a daily
basis based on the number of sales each day; the current day's
per-share dividends are made available to all users. In other
embodiments, other periods of time, such as weekly, monthly, or
quarterly are used in the calculation of dividends. As is the case
with real stocks, dividends are subject to change over time.
[0404] In some embodiments, the system is not strictly tied to the
model of the stock market and securities, but uses other means for
the same effect.
[0405] In particular, what is needed is a means to enable people
who accurately predict which new items will be highly valued to
profit by those predictions.
[0406] One such means is to enable people to buy the new item at a
reduced price compared to the likely later price, so that those
items can be resold later. This is the same principle that guides
people who buy particular baseball cards today in the hopes of
selling them in the future at a profit if those particular cards
become more valuable in the future.
[0407] For example, in one embodiment relating to music, means are
presented whereby the publisher of a new song can specify a price
to sell the song, where such price may only be available for a
limited amount of time. Subsequently, users are presented with the
means to buy a number of copies of the song. This may occur by
downloading multiple copies of the song. Alternatively and
equivalently, only one digital recording may be downloaded, but
with a contractual right specifying the number of "copies" it
represents by giving the user the right to make that number of
copies, including the right to resell those copies to other people.
The contract can be presented online and signed with a digital
signature. Alternatively and still equivalently, no copy needs to
be downloaded at all. Instead, the number of "copies" purchased by
the user is kept track of by the system. In many such embodiments,
the system subsequently presents means for other users to purchase
the copies from the original purchaser. The system is an agent for
the selling of the copies, similar to a store which sells used
CD's, except that in the embodiment being discussed the recordings
are virtual: only one physical copy need be stored on the magnetic
media of the server (or, it can even be stored on another machine),
and the server keeps track of who owns how many "copies" in its
database.
[0408] Of course, this aspect of the invention should not be
construed as being limited to music; any digital item can be
equivalently handled by an embodiment of the invention. Examples
include video recordings, electronic books, articles, etc. Further
examples include such items as concert tickets, where the tickets
can be purchased for new artists at a reduced price for concerts
that haven't been scheduled yet, such that the tickets are good for
any concert and are sold by the system before other tickets. The
tickets may be kept only in digital form for concerts broadcast
over the Internet that may only be seen by ticketholders, or the
system may provide means for tickets to be displayed on the screen
and/or printed for concerts that the ticketholder can physically
attend.
[0409] In some embodiments, the system resells any of the
earlier-purchased item that the owner desires to resell before it
sells "new" items. That is, if a user (Joe) likes a new artist and
buys 1,000 copies of his recording in the hope of reselling them
later for more money, the system is required to resell whatever
copies Joe makes available for resale before it can sell other
full-price copies of the recording. Otherwise, Joe may not be able
to efficiently resell his copies, and the ability of the system to
motivate Joe and others like him to identify high-quality new items
will be compromised. To facilitate this requirement, means are
provided whereby Joe can input the number of copies he wants to
resell at the "official price" at the current time. In general
embodiments of this invention generally provide an order in which
items are resold by the system when a number of users are trying to
resell items at the same time; the preferred ordering is that in
which the most profit is made per sale, however other orderings are
possible.
[0410] Other embodiments provide no such structure and instead rely
on market processes to accomplish similar aims. For example, there
is no requirement that any particular copies be sold ahead of other
copies. However, it is understood that whoever bought copies for
the lowest price is in the best position to slightly underbid the
rest of the market while still making a profit. So, no formal
mechanism need be included to enable him to sell his copies. In
must such embodiments, any owner of one or more copies can offer to
sell one or more of them at any time for any price. These offerings
are posted (in most cases, the currently offered copies are listed
together in an area, such as a Web page or set of Web pages, that
is dedicated to the song in question. Such listings are useful when
a user would like to buy a number of copies, so that he can
eventually resell them himself. One exemplary user interface lists
offerings in lots, where all copies in a lot have the same price.
Input means is provided where the user can input the number of
copies he wants to acquire, and an output is generated showing the
total cost and the average cost per copy, where the copies are
purchased from the least-expensive lots such that one lot may only
be partially used, but all copies in less-expensive lots are
completely sold out.
[0411] In some embodiments the user can specify a maximum average
price he would like to pay, and the system will automatically
reserve the least-expensive collection of copies available, compare
the average price maximum average price, and: a) if the average
price is less than or equal to the maximum, the purchase is
completed; or b) if the average price is more than the maximum, the
reserved copies are released and the user is informed that the
desired number of copies could not be purchased at the desired
average price.
[0412] However, in cases where a consumer simply wants to buy one
copy, a simple interface may be provided in which only the
lowest-priced offering is displayed. In cases where the items are
completely digital, there is no difference between copies, and
therefore no reason to show any but the lowest-priced offering.
[0413] In some embodiments a user who wishes to sell an item is
presented with a screen showing (at least) the least expensive lots
currently offered for sale, so that he can consider them in
deciding how to price his offering. In further embodiments, for the
sake of convenience, the user is enabled to input the price and
number of his own items he wants to sell on the same screen as the
one where he views the lots that other people are selling.
[0414] Other embodiments may create an auction marketplace modeled
closely after eBay and other Web-based auction sites.
[0415] Some embodiments accomplish these aims by providing "Right
To Download" (RTD) certificates, which may be completely
electronic. In some such embodiments, there is a database, which
may be distributed, which stores RTD objects. These objects contain
an owner ID, a secure identifier such as a password (which may be
encrypted), an identifier of the digital item represented by the
object (for instance, a particular musical recording in LiquidAudio
format), and the current asking price (and may have other
information as well such as the entire purchase history of the
object).
[0416] An owner of such an RTD can sell it at any time. In one
embodiment, there is a Web site associated with the database. The
owner of an RTD can use the user interface on the Web site to
indicate he would like to sell one or more of his RTD's for a
particular item. This causes that number of RTD's, owned by that
user and associated with the item in question, to go into a "for
sale" state. For instance, this may be accomplished by setting a
flag in the RTD object in the database. In various further
embodiments, these items can either go on sale at a price chosen by
the owner, or be auctioned.
[0417] In other embodiments, RTD's can be bought and sold by other
means. For instance, each RTD may be given a unique identifier. A
transaction may be created in which credits or money is transferred
from the buyer to the seller causing the owner information in the
RTD object in the database to change from the seller's information
to the buyer's information. One example of a way to take advantage
of this type of transfer would be on an auction site such as eBay.
A seller can auction off an RTD for a particular item by means of
the auction site's standard mechanisms. When the auction is
complete, the seller goes to the Web site set up for the purpose of
completing such transactions and sets an attribute of the RTD
object such that the object goes into a "sale in progress" state.
The seller also enters the agreed-upon amount into another
attribute of the object. He then sends the RTD object's unique
identifier to the buyer.
[0418] The buyer can then log onto the same Web site, at which he
enters the unique identifier of the RTD object and his credit card
information. The site presents a confirmation that he wishes to buy
the right to download at the agreed-upon price (stored in the
object) and if he agrees to the confirmation, the RTD is now
his.
[0419] When the owner of an RTD actually downloads the item
associated with the RTD, an attribute is modified such that the RTD
goes into a "download completed" state, indicating that no more
downloads may be made using that RTD. This makes the RTD
valueless.
[0420] In some embodiments, downloads may be made directly from the
Web site whose purpose it is to complete such transactions.
However, using interprocess communications mechanisms such as SOAP,
a separate site can initiate the download by accepting the unique
identifier of the RTD object, and checking with an RTD server
process (which usually exists on another machine on the Internet)
to see that an item exists with that identifier and that it is not
in the "download completed" state.
[0421] When a person buys RTD's, in some embodiments he can
immediately make them available for sale at another price. Of
course, assuming the other price is higher, as will certainly be
the usual case, it's unlikely that there will be any buyers,
because RTD's for lower prices will normally be queued ahead of the
RTD's in question. But when the time comes that there are no
lower-priced RTD's, the ones in question will automatically be
sold.
[0422] The concept behind the RTD also applies beyond the world of
downloads. CD's may be purchased and held, or even pre-purchased
before being made, and that ownership can be transferred using the
same principles as described for RTD's. This can be considered to
be a variant of the right-to-download concept appropriately called
the right-to-buy (RTB). It is exactly the same idea, but instead of
allowing a download to occur, the physical shipment of the item is
allowed to occur. While the focus in this specification is on RTD's
because of that concept's intrinsic appeal in an Internet
environment, the invention should not be construed as being
restricted to RTD's to the exclusion of RTB's.
[0423] Note that the "buying" aspect of an RTB can involve a
transfer of no money, in which case the object can be seen more as
a right-to-be-shipped an item rather than a right-to-buy. But in
many cases a non-zero purchase amount for an item will make more
sense. For instance, there are certain fixed manufacturing and
shipping costs associated with an item. These can be represented in
the buy-price encoded into the RTB. Thus, an RTB itself might cost
$0.10 at first, and as the associated item attains real demand,
subsequent resales of RTB might go up to $1.00 or even $10.00 or
more. However the fixed purchase price may be $2.00, covering the
fixed costs of manufacturing and shipping the item.
[0424] Thus the last person to purchase the RTB might pay $10.00
for it, and then another $2.00 to actually complete the transaction
and cause the item to be shipped.
[0425] In other embodiments a "right-to-experience" is sold rather
than an RTD or RTB. By "experience," we might mean "view" for a
visual work, "hear" for a musical work, etc. It is the act of
experiencing a work rather than acquiring it. For instance, in some
scenarios, works might be freely downloaded to a user's hard drive,
but a fee is charged for experiencing them. This may be a
per-experience fee (i.e., per-listen, per-view, etc.) or it may be
a fixed fee, for instance, a monthly fee for unlimited listens.
[0426] One fixed fee scenario in the music realm would be a service
that gave users the right to download and listen to any music they
want to download and listen to, as frequently as they want (or
perhaps limited by some maximum number of downloads or other
constraint) for a monthly fee. For instance, one can imagine a
service such as Napster charging $20 per month for use, enabling
the user to have access to any music he can find in the Napster
network. A player such as RealJukeBox could easily be modified to
work with Napster to count the playings, and provide those counts
to a central server.
[0427] In such cases, the organization providing the fixed-fee
service may nevertheless be required to pay proportional to the
number of experiences. For instance, radio stations provide music
for a fixed fee (nothing) but are required to pay royalties based
on audience size and the number of playings, or on music-related
revenues. Then, one can imagine a system in which money is paid to
the owners or producers of recordings based on the proportion of
the overall hearings each recording attracts. For instance, user A
and user B might each be playing $20 per month for an "unlimited"
music service. However, user A might listen to 1000 songs (not
necessarily unique) per month, whereas user B might listen to 100
songs per month. If, in each case, 5% of listening time goes to
Rolling Stones songs, user A is playing Rolling Stones songs many
more times per day than user B, while not paying any more
money.
[0428] This is not incompatible with artists, song owners, or
"right-to-experience" (RTE) owners getting a fixed fee for each
hearing. It simply means that the organization providing the
fixed-fee service will make more money from user B (it might even
lose money by user A, but the amounts overall will be such that at
least some fixed-fee service providers will be able to make money
overall by providing these services).
[0429] Some embodiments of the present invention are thus geared
toward providing a marketplace for RTE's, similar to the
marketplaces for RTD's and RTB's. In this case, however, the buyers
may be organizations providing fixed-fee services rather than the
end-consumers themselves. Such organizations would be motivated to
take advantage of these markets in order to buy bearings more
cheaply by buying hearings associated with lesser-known artists,
whose market prices for RTE's are probably less. These
organizations would then be motivated to market such artists to
their users in order to get users to listen to those artists rather
than the more currently popular artists; in doing so, they lessen
their costs while, ideally, maintaining their revenue stream.
(Alternatively, fixed-fee services can be set up that do not keep
track of the number of hearings but only of the number of downloads
of each recording. In such a case, the RTD market is used similarly
to save money.) In some embodiments, no traditional "download"
takes place, as in the case of "Internet radio stations" which
broadcast streaming audio such that if the user hears a song twice,
it has been streamed twice. In such a case the RTE paradigm is a
good match.
[0430] In preferred RTE embodiments, RTE's are frequently bought in
lots, such as 10,000 RTE's. In some such embodiments an interface
is presented to buyers allowing them to request a number of RTE's
of their choosing, for instance, above a certain minimum.
[0431] In some embodiments, RTD's may be sold at various Web sites
which communicate with the system which stores the actual RTD
database objects by means of interprocess communication facilities
such as SOAP, Java RMI, or other similar technologies. In some such
embodiments, there a listing is made available by those means which
lists identifiers of the items for which RTD's currently exist; the
system can also preferably provide descriptive information about
those items such as titles, performers or other kinds of
information which pertains to the particular type of item. Another
interface is provided by means of which an RTD for a particular
kind of item is requested. When an RTD is made available in this
way, it is "locked," and other requests for RTD's will result in
other RTD's being made available. The unlocked RTD with the lowest
price is made available next. (In other embodiments, market-making
mechanisms are used to provide a "current price" for the RTD's
associated with an item; this enables all sites currently selling
an RTD's associated with an item to sell them for the same current
price.)
[0432] In some further embodiments, sites can choose to sell RTD's
together with an automatic invocation of the RTD. In such cases,
the user does not need to know anything about the concept of an
RTD; all he knows is that the item is being made available at a low
price; he pays the price by credit cards or other techniques and
the item is immediately downloaded. In the case of an RTB, the
price presented to the user may be the price of the RTB plus the
associated fixed per-item cost, if any; so the user only sees the
price he needs to pay in order to have the item shipped to him.
Again, he may be completely unaware of the underlying RTB
mechanism.
[0433] It should not be construed that the RTD (RTB) concept is
limited to the implementation details spelled out here. The
fundamental idea is that there is a database with records
indicating the state of each RTD, and there are one or more
identification fields by which a the data for a particular RTD can
be located in the database along with permission granted to modify
it. This identification fields typically but not exclusively
involve a unique identifier for the RTD records, as well as a
unique identifier for the owning user, and a password for that
user. RTD's can be transferred between people at different prices
at different times. One RTD can represent one right to download (or
buy); in some embodiments, rights to download or buy multiple items
may be incorporated into one set of RTD records, although such an
approach might be more complex than the one-RTD-per-item
approach.
[0434] In some embodiments, RTD's are used to determine the
prescience of the owner of those RTD's. For instance, if we take
the average of the last few sales of the RTD's associated with an
item as an indicator of the current market value of such RTD's,
then the price at which the owner originally purchased those RTD's
can be compared with their market value, and an average change per
RTD calculated. This average can also include the change in price
between purchase and sale for RTD's he no longer owns. This
average, when considered alongside with the number of RTD's the
user has purchased, is a good indicator of his prescience.
[0435] For instance, one way of combining these numbers is a + S b
+ N , ##EQU13## where S is the sum of the changes in value for
every RTD the user has purchased, N is the number of such item's,
and a and b are chosen such that a/b is the average change in RTD
value across the entire population (excluding the user under
consideration), and b is chosen to be consistent with the amount of
evidence one RTD gives regarding a user's prescience. This can be
determined using historical data once the system has such data
using an optimization algorithm. As an example of how to do this,
the system can "hide" one randomly chosen RTD owned by each user in
the system. Then, for a particular "guess" regarding the optimal
value of b, we calculate the user's prescience, ignoring the hidden
RTD's. We then calculate the average difference between the
calculated prescience and the actual change in value of the hidden
RTD's. We then use an optimization algorithm to find the value of b
that minimizes this average difference. For example, if the
optimization strategy is genetic algorithms, and d is the average
difference, then an appropriate fitness measure for a particular
value of b would be 1/d; the genetic algorithm would find the value
of b with the maximum fitness.
[0436] This prescience measure can then be used as the determinant
of the user's ability to place items in a publicity list or
external valuation list.
[0437] In preferred embodiments, there is no restriction against a
user being able to buy copies at different prices at different
times. For instance, when a high-quality recording from a new
artist is first made available, it may be able to command only a
very low price. As its popularity grows, it may be able to command
incrementally higher prices. Users who are interested in profiting
from this ongoing growth in value may wish to buy recordings at the
current intermediate price in order to resell at an anticipated
higher price later.
[0438] In some embodiments, the publisher of items may guarantee
that he will never sell items for less than a particular price,
thus increasing the security of potential early buyers. In some
such cases, the guarantee may apply only to copies other than the
ones being sold at the time; i.e. a user may buy 1,000 copies for
price X while being guaranteed that other copies will not be sold
for less than twice that price.
[0439] Taking this idea further, in most (preferred) embodiments
appropriate constraints will be provided so that, for instance, the
price of each download isn't forced to zero. (This would happen if
the number of downloads being offered on the trading market was
greater than the number of people who were interested in the
recording, and there were no lower limits on pricing. Then, holders
of downloads would all try to underprice each other, forcing the
price toward zero.)
[0440] We therefore need a market control mechanism--a way to make
sure that the market price of RTD's (or similar items--for
convenience the language will discuss RTD's but also applies to the
variants discussed above such as RTB's) purchased by speculators
will increase over time, enabling the speculators to make a
profit.
[0441] The first such mechanism to be discussed is to make a
limited number of items available at a low price at the time of a
song's release and to sell all other RTD's at a fixed, much higher
price.
[0442] For instance, assume a new, unknown band wants to use the
market to increase the speed with which its new song gets known. It
may then make 10,000 RTD's available at $1.00 per RTD, and promise
to charge $12.00 (in some embodiments, this would be a minimum
promised price, not a fixed price) for all subsequent RTD's.
[0443] Then speculators may buy all 10,000 initial RTD's. If the
song does indeed become very popular, those speculators who wait
until then to sell their RTD's will be able to charge very near to
$12.00 for each one. Such speculators therefore will make nearly 12
times their initial investment.
[0444] Setting a minimum price for RTD's sold to people other than
the initial speculators is one constraint mechanism which can be
used to cause the market price for the speculator's copies to rise.
Another way is to limit the number of RTD's that are available at
any given time. The approach to doing this that I will describe is
from a verbal suggestion of Robin Hanson. However, it must be
stressed that the invention is not limited to any particular market
constraint mechanism; the examples being given here are examples
only. Under this approach, an initial number of RTD's is made
available by the owner of the item at a low price. Many or all of
these are purchased by speculators. Once these have all been
purchased, if it seems that there is a larger market for the item,
the owners of RTD's for the song can vote on the question of
whether more RTD's should be generated. If the decision is "Yes",
than, in some embodiments, each RTD becomes two RTD's. In others,
means are provided for RTD owners to also vote on the number of new
RTD's to be generated. For instance, different proposed numbers can
be voted on, and the one with the most votes wins. By thus allowing
the number of RTD's to be increased in a limited fashion, more are
made available for people to buy (and speculators to profit from)
while it is not true that so many are created that the market price
is driven toward zero.
[0445] Of course, all this may be accomplished via a Web or other
network computer interface where techniques in common use and
obvious to any competent practitioner are used to present RTD
purchasing, voting, and other necessary interfaces.
[0446] In some embodiments, only profits obtained by resale may be
used for the various publicity mechanisms such as a publicity list,
or those profits plus some limited amounts of other funds may be
used.
[0447] In some embodiments, the number of copies made available at
a given price may be limited.
[0448] In some markets, the value of an item may be very much
related to personal tastes. For instance, at the time of this
writing far fewer people in the United States are interested in
instrumental dulcimer music than in mainstream pop. A publisher who
wishes to maximize the value he will derive from his items needs to
be very cognizant of the size of his market. For instance, if a
particular genre has 10,000 people interested in it, and the
publisher makes 10,000 copies of an item by a new artist available
at a very low introductory price, there won't be anyone left to buy
the item later at full price when the artist becomes better-known.
Thus, if the artist eventually becomes highly respected within the
genre, a significant sum of money will have been "left on the
table."
[0449] For the sake of such markets where these issues are a
concern, means are provided for helping users understand the size
of their potential markets.
[0450] For instance, allowing the user to specify a particular
recording in the genre and see how many copies of have been sold.
The utility of this may be increased by also showing this number as
a percentage of the overall population.
[0451] In embodiments which break the market into various
categories (such as musical genres), the average number of sales of
songs in the genre may be shown; additionally it may be displayed
as a percentage of the average number of sales in total.
[0452] Another method for determining the potential market size for
a song is provided in some other embodiments (this general market
size can be automatically and tentatively applied to all songs from
a given artist). Means is provided for the publisher to specify
artists who should appeal to similar tastes as the current artist
or song. In preferred embodiments, they are listed in order of
similarity; that is, the most similar artist is listed first, the
next most similar artist second, etc Then the actual market sizes
for these artists are calculated and averaged together, with the
numbers for the most similar artists being given the most
weight.
[0453] Various embodiments of the present invention use different
means for paying any money due to users. In some, in order to
minimize overhead, dividends are only paid on a monthly, quarterly,
or annual basis, or only after they have accumulated to being
greater than some figure. In some embodiments payments are made
directly to checking accounts; in some checks are mailed; in some,
the payments are credited to credit cards; in others, other means
are used.
[0454] In some embodiments, purchases of stocks or items can be
made automatically under certain rules. For instance, a user may
want to buy 1,000 copies of each new work by a particular musician
as long as he can get them for less than a particular price. Input
means are provided whereby the user can specify such actions.
[0455] In some embodiments, means are provided for otherwise
unrelated Web sites (or other entities facilitating networked
computing) to participate in the market mechanism; in preferred
such embodiments, they are enabled to profit from doing so. For
instance, in one such embodiment, SOAP is used to communicate the
necessary information to enable other Web sites to provide the
inputs and outputs necessary to facilitate transactions in the
market, such as display of items available to buy shares in, the
price per share, current dividends, etc. In preferred such
embodiments, an interface is provided whereby such an entity can
tell the market service when an item sale is made, for instance,
when a song is paid for via a Liquid Audio download; this interface
uses similar means such as SOAP. In most embodiments, entities that
help facilitate the market in this manner receive a share of the
profits.
[0456] The "market price" of an item may be displayed for users to
see. In embodiments which use market-making mechanisms, the current
market price as determined by that mechanism may be used. In some
embodiments, the current lowest offering price of an item is used
as its "market price." In others, the most recently completed sale
price may be used. In some embodiments, the market-making mechanism
may be willing to buy at one price and sell at a slightly higher
price; thus, "bid" and "asked" prices can be presented. Other
variations are possible and also fall within the scope of the
present invention. In some embodiments, the market price
information, however it may be determined according to the
particular embodiment, is made available by an interprocess
communications mechanism such as XML-RPC or SOAP. This information
may be used as one measure of the likely quality of an item which
may be usable by other services. For instance, Web search engines
use many factors in deciding how high to rank various Web pages
that may contain the key words the user is searching for. One such
factor, in certain advanced search engines such as Google, is
counting the number of links to each page. Another factor, to be
used in a similar way, could be the current market valuation of
each page.
[0457] One possibility that must be contended with when trying to
use the market value of an item as an indicator of its quality is
the potential for fraud; that is, users who make certain trades
expressly for the purpose of making an item seem more valuable than
it really is. For instance, the creator of a musical item could,
under a set of pseudonyms, purchase all the RTD's made available at
the beginning of the item's market availability, and then buy and
sell the RTD's using those pseudonyms at any price; if, for
instance the last sale of the item were used in some context as a
measure of its quality, then he could make a sale between two of
his pseudonyms at a high price, thus creating a spurious measure of
high quality.
[0458] Thus, in embodiments where market value is used as an
indicator of quality, certain mechanisms may be used to hide (or
denote as unreliable) the market value for items where that market
value is questionable. For instance, if the number of entities
trading an item is smaller than some fixed number, the market value
may be hidden or made unavailable through the interprocess
communication mechanism; this is because the more entities involved
in trading an item, the less likely it is that those entities are
all pseudonyms or people in collaboration with the stakeholder in
the item. Another mechanism has two key parts; first, there must be
a transaction cost, possibly paid to the operator of the market,
and second, there is a requirement, for a market value to be
considered valid, of a certain number of daily trades. Under this
mechanism, people who try to manipulate the market will find it
costly, since they will be paying the transaction fees but not
realizing any upside from their trades; whereas legitimate users
who are prescient about which items will go up in value will be
able to make money from their trades. This will encourage
legitimate, prescient users to continue trading while discouraging
fraud. Through real-world trial-and-error, an appropriate balance
of the number of required trades and the cost per trade can be
determined. As elsewhere in this specification, the details are for
purposes of example only and should not be considered to limit the
scope of the invention. Like other concepts described in this
specification, the present one should not be considered to be
useful only in the context of the other aspects of this
specification.
[0459] One possible use for the invention is a system that stores
product suggestions, and uses the described mechanism to bring the
most attention to the most important suggestions, and to enable the
people that make and publicize those suggestions to be highlighted
to be rewarded. To that effect, an set of screens should be
provided whereby manufacturers can set up accounts where they can
exchange credits earned in the system for actual money.
Alternatively, they can make their products available for credits.
The interface could optionally enable them to only compensate
credits earned for suggestions relating to their own products; the
system would thus keep track of the source of all credits. This
scheme may be appropriate in other situations as well; for
instance, a site could be devoted to finding human interest stories
for newspapers; the newspapers could pay people for credits earned
in helping them find stories appropriate to their particular
newspaper. In some embodiments, there are separate areas for
discussing the products of different companies. In some such
embodiments, users or management of the system can create areas for
various companies without the involvement of those companies.
[0460] In cases where product improvement ideas are the items, the
value, in some embodiments, is based on the proportion between the
value of the feature and the amount of work it will take to make.
For instance, an easy fix of a moderately-important bug might float
to the top of the list, because this proportion would be high. To
make this happen, text descriptions and help information says that
that is what the value should be based on. This makes sense because
it will form the basis of the company's decision whether or not to
act on a suggestion. The external value comes from the question of
whether a company acts on the suggestion; thus there must be a way
for the company to input information on whether this has occurred.
(This could involve a logon ID for each company, or preferably
different logon ID's for various authorized representatives of such
companies; their records would include their company identifiers.
Certain individuals are authorized to indicate that a another
person is authorized by the company in question.) In this scenario,
some embodiments keep the external value undermined until official
company action takes places: either the company fixes the bug or
dismisses the item as something it will never address. (The same
mechanism will work for feature suggestions as well.) If the item
is dismissed, the external value is 0. If the item is accepted (bug
fixed, or new suggestion integrated into the product) the value is
some number which may be the same for all accepted items. Once the
outcome is determined, all shares have the appropriate value given
by the external measure.
Miscellaneous Notes
[0461] FAQ-Generating Application:
[0462] Investors share ad revenues for FAQ items. The external
measure incorporates not only the number of access, but the
percentage of times the answer was not satisfactory. Investors can
email the answer owner to change its place in the hierarchy or to
change its keywords or the weights on those keywords. If the owner
isn't responsive, ownership switches to somebody else who gets
equity, although the original owner also retains some equity.
[0463] Mix a certain number of publicity list entries with
popularity list, invisibly.
[0464] For weblogs, make ultimate valuation tied to visits per day.
Current should be based on expectation of that.
[0465] The invention may be used to facilitate a task force to
solve a particular problem. In some embodiments users "buy" their
way onto a particular task force by means of credits. This ensures
that participants are more highly motivated, increasing the
"signal-to-noise" ratio.
[0466] Have sign-off sheet with specific punishment if they use
automated or manual means to sabotage the popularity counters for
sales, downloads, weblogs, etc.
[0467] Publicity list also links to reviews. Reviews are qualified
in that the reviewer only suggests you will agree with his review
if you are within a certain distance of his taste, where that is
given, for example, in percentage of the population. One
possibility: look at number of MP3's on disk to determine your
taste--download needs to be able to scan them.
[0468] Publicity list includes links to free samples.
[0469] Offer people a way to make suggestions to the author to
improve a brainstorming idea after they've invested in it; deal
with the question of whether it makes sense to do that or create a
new item . . . they should want to help the author first, and if
that fails, try on their own.
[0470] Enable any individual to create a fund.
[0471] For product improvement ideas, in some embodiments the
company orders them in value or assigns a value after a certain
amount of time or only for those reaching a certain point.
[0472] In embodiments powering a "mega FAQ," enable people to
suggest keywords for entries to make them more valuable--also have
the value of an entry related not only to the number of times the
user's query is satisfied, but also reduced by the number of wrong
hits to an entry. This motivates people to make their queries dead
on. They can actually associate questions such as "how many so and
so in a so and so".
[0473] Preferred embodiments enable people to give their permission
("opt-in") for feeding their info into an engine which allows
musicians to choose where their work will be sent to for initial
publicity. People with great track records with regard to
prescience, for instance, as determined by points earned, are the
most appropriate recipients. But some people will not want their
info made public due to privacy concerns.
[0474] In some embodiments, the number of credits earned by a
person is considered to be an indicator of the degree to which he
has contributed positively to the community. For instance, on a
free-of-charge MP3 site which aims at connecting artists with their
ideal audiences, people who earn credits through investing are
people who were prescient in identifying the best new recordings
and artists before the rest of the community. In many cases, such
people helped to popularize the artists by posting to the publicity
list or by other means (certainly, anyone who invests in a
particular item is motivated to help popularize it, and therefore
it can be assumed that this occurs a certain amount of the time).
In embodiments where credits are not convertible to money, items,
or other usual means of compensation, value may still be found in
credits by publicizing the number of credits earned by each person
(or the rankings of credits or other forms of recognition based on
the number of credits). Such publicity should be made in the
context of a description of the fact that such publicity is
recognition for adding value to the community by identifying the
best items early. For instance, one simple embodiment lists the Top
Ten Contributors to the Community by choosing the 10 people with
the greatest number of credits and listing them in order of credits
(or alphabetically, or some other order). Thus, this recognition of
being a real contributor to the community is the ultimate currency
represented by credits.
[0475] Instead of buying "shares" in an item, the same effect can
be achieved in many embodiments by selling "percentages" of the
item or its related profits.
REFERENCES
[0476] Bugzilla: bugzilla.mozilla.org/ [0477] SOAP specification:
http://static.userland.com/xmlRpcCom/soap/SOAPv11.htm [0478]
http://www.redherring.com/cod/2000/0125.html [0479] Google:
www.google.com [0480] Napster: www.napster.com [0481] Gnutella:
gnutella.wego.com [0482] Papers related to "idea futures" by Robin
Hanson at http://hanson.gmu.edu/ifpubs.html#Hanson
[0483] The entire contents of U.S. patent application Ser. No.
09/714,789, filed, Nov. 16, 2000, are incorporated herein by
reference, including the specifications, drawings, and abstracts,
are hereby incorporated herein by reference.
BACKGROUND OF THE INVENTION
[0484] 1. Field of the Invention
[0485] The present invention relates to business methods, and to
techniques and systems for financing items through future sales
rights. The invention has particular advantages when used to
provide sales options for copyrightable material such as
entertainment recordings. The invention further relates to a system
to determine quality through reselling items. The context for this
invention is network-connected computer systems which allow a
number of individuals to interact with a central system for
carrying out these sales.
[0486] 2. Background
[0487] In the entertainment industry, manufacturers and
distributors are faced with fixed costs of manufacture, and
distribution, regardless of quantity or popularity of the
entertainment product. This means that they must make an estimation
of the future popularity and sales of the particular item. While in
some instances the manufacturer and distributor have accurate
predictive data, it is particularly difficult to predict the degree
of acceptance and market success most entertainment items will
have.
[0488] Regardless, there are generally people who have a
significant degree of understanding of particular entertainment
markets, and who can judge the potential success of a particular
item. These people may be willing to provide financing for
entertainment products in the particular entertainment markets. To
the extent that these people can provide financing and can provide
good predictions of the performance of entertainment products in
the marketplace it would be desired to create a market structure
which allows their knowledge and expertise to be used financing
create an economic incentive to manufacture or distribute
entertainment, copyrightable or other products.
[0489] This invention is intended to hasten the identification of
high-quality items by enabling those who are particularly good at
identifying them to make a profit from doing so. We do not
distinguish between physical objects such as paintings and digital
objects such as MP3 files except as noted below.
[0490] We assume the existence of objects such as musical
recordings which may or may not have value to a particular
community. In the digital world, an MP3 may be in a genre that only
a small subset of the population is interested in. But, as a
separate matter from the genre, it may or may not have
quality--quality that would motivate people interested in that
genre to want to hear it.
BRIEF SUMMARY OF THE INVENTION
[0491] In accordance with the present invention, a method of
optimizing valuations of items in a market includes establishing
values for the items and providing a sequence of the items for
which speculators may invest. The items are then sold to consumers
of the items, or rights to the items are transferred to consumers.
The speculators who bought the income rights to the items are then
provided with income in accordance with the income rights for the
particular ones of the items sold to the consumers or for the
rights transferred to consumers.
BRIEF DESCRIPTION OF THE DRAWINGS
[0492] FIG. 1 is a flowchart showing the transfer of funds
according to the present invention;
[0493] FIG. 2 is a diagram showing database tables in accordance
with one embodiment of the present invention;
[0494] FIG. 3 is a flowchart showing original owner creating RT's
in accordance with one embodiment of the present invention;
[0495] FIG. 4 is a flowchart showing a speculator buying RT's in
accordance with one embodiment of the present invention; and
[0496] FIG. 5 is a flowchart showing a consumer finalizing an RT in
accordance with one embodiment of the present invention.
DETAILED DESCRIPTION OF THE INVENTION
[0497] For the sake of simplicity, we will focus on one particular
example--MP3 files. But it must not be construed that this
invention is limited to such files; it equally well applies to
paintings, books, CD's and other objects which consumers may find
appealing or unappealing.
[0498] A problem to be addressed is the fact that there are a huge
number of MP3's, most of them of very low quality regardless of the
genre they may happen to be in. It takes time and effort to listen
to many of these recordings in order to find the gems. Those who
perform that service--who we shall refer to as "scouts" (or
speculators)--should therefore have the opportunity to be rewarded
for it.
[0499] One aspect of this invention is that the entity owning the
MP3 is given the ability to sell a certain number of "rights" at a
low price with the expectation that these rights will be worth more
later; the scouts would be rights to the MP3's they think will be
popular in the future with the expectation of reselling them later
for a profit.
[0500] The meaning of the term "rights" varies from embodiment to
embodiment. Several variants that fall within the scope of the
invention are listed here, although this list is not meant to be
exhaustive:
[0501] A right to download (RTD): Applicable for MP3 files or other
digital works. The scout buys the RTD and resells it later at a
higher price to a consumer, who actually does the download. A
database keeps track of who owns the RTD at each point in time;
sales are registered in the database.
[0502] A right to experience (RTE): This is applicable in some
circumstances to MP3 files or other digital works. For instance, in
the context of a subscriber service, consumers do not necessarily
own MP3's, but pay a fee for the right to play them. A user can
play any available song at any time, and it may be streamed from a
remote server or it may exist on the local audio or computer
system. The subscriber may not pay for each play of a given song,
but in the current scenario, some entity does--most likely, the
organization providing the subscription service. Since each play
therefore has value, scouts would be well-advised to buy RTE's at a
low price before an artist is well-known, and resell them
later--often directly to subscription services--at a higher price
when a demand has been built. Thus, subscription services buy RTE's
according to how many times users play various songs.
[0503] A right-to-buy (RTB): This is applicable to physical objects
such as CD's. Scouts buy the right-to-buy a CD at a certain price
(which may be $0; i.e., the purchase of an RTB may be the only
necessary purchase. It may be non-zero to cover such aspects as
shipping and handling, which would be fixed for the lifetime of an
RTB). Scouts then resell the RTB to consumers at a higher price
after a demand is established.
[0504] For short, we will call such variants RT's that is, "rights
to . . . "
[0505] In each case, there is a "finalization event." For RTD's
this occurs when the object is downloaded. For RTE's, it occurs
when the object is experienced (for instance, a listener hears a
streamed audio file once). For RTB's, it occurs when the physical
object is bought (and/or shipped).
[0506] For purposes of example, we will focus on RTD's relative to
songs, but except where noted the concepts apply equivalently to
RTB's and RTE's. Also, they apply equally well to various formats
in which digital property can be encoded; some formats in fact may
be superior to MP3's for profitable transactions due to built-in
copy protection features; Liquid Audio is an example of a company
marketing such technology.
[0507] Suppose a scout buys 1,000 RT's for a particular song at
$0.10 each with the expectation of selling them at a later time for
$1 each. But suppose the owner of the song makes a large number of
RT's available through a number of channels at $1 each. This could
very significantly slow the rate of sales of the scout's RT's. In
fact, if more RT's are made available than there is a market for
RT's, many of the scout's RT's might never be sold. This means that
after the scout buys RT's, his fortunes are tied to subsequent
management decisions on the part of the original owners. So his
profit is determined not only by his own prescience in determining
which new recordings are likely to become popular later, but also
by the unpredictable decisions of the owners.
[0508] One factor making it hard to avoid this kind of problem is
that the market size for a particular song can not be exactly known
in advance. This motivates the owner to make as many copies
available as possible through as many channels as possible.
Moreover, a sale made by a scout means the owner gets $0.10 for
each copy; a direct sale means the owner gets $1 for each copy.
[0509] So the fortunes of the scouts and the fortunes of the owner
are at odds.
[0510] This creates the possibility for trouble. Even if there is a
contract stating that RTD's after the first 1,000 will be sold for
$1 each, the fact is that if the owner violates the contract, the
scouts could lose money and be faced with the expensive prospect of
suing for damages.
[0511] So, ideally, there should be a technique for eliminating
this problem. Such a technique is a key aspect of the
invention.
[0512] The solution provided by the present invention is to
finalize the RT's in sequence. For instance, if a scout buys the
first 1,000 RTD's, he will supply the first 1,000 RTD's which are
downloaded. The original owner may way to flood the market with
direct-sale RTD's, but he cannot be finalized until the scout
finalizes his RTD's.
[0513] This mechanism even protects against the extreme case of an
owner selling RT's at $0.10 each and then subsequently
direct-selling the RT's for an even lower price (which he might
want to do, for instance, in order to use that song as a
loss-leader, building popularity, so that he can sell other songs
more quickly later).
[0514] If the owner then says he is going to flood the market with
$0.05 RTD's, people will want to buy them at that price--but they
won't be available. If the scout was right in his belief that
demand would exist such that consumers would eventually want to pay
a higher price, then the consumers who are most eager to obtain the
RTD's will buy them at the higher price, since the alternative is
not to have them at all, at least for a very long time. The scout,
knowing this, can wait as long as necessary to sell his RTD's at
the higher price.
[0515] This mechanism therefore protects the scouts against
decisions by the original owners that could otherwise have a
negative effect on the scouts. It applies equally well to RTE's,
RTB's and other equivalent forms of RT's.
[0516] Any of a number of market mechanisms may be used to carry
out the sales. For instance, blocks of some fixed number of RT's
may be auctioned on eBay. (In fact, eBay has recently announced
that it will make software mechanisms available for 3rd-party
companies to set up auctions without direct human intervention.)
Or, market-maker software may be provided emulating the
market-making techniques used in various stock markets. In
preferred embodiments, the original owner controls how many RT's
are made available for sale to the scouts at any point in time.
[0517] In preferred embodiments, RT's may be sold in any order
until the finalization event occurs, at which time they are removed
from the marketplace. That is, for example, and RTD for the
10,000th download may be purchased before the RTD for the 5,000th
download is purchased. In preferred embodiments, there is a free
market for selling RT's. At any time, scouts may buy them or such
organizations as retail stores can by them, without
distinction.
[0518] (In some embodiments, consumers may buy them too, and
subsequently trigger the finalization event for their own use.
Interfaces are provided for such purposes. For instance, in the
case of RTD's, in some such embodiments each RTD has a unique
identification number. When a consumer purchases an RTD, the ID is
presented to him by such means as a Web interface or email. For
instance, if a consumer purchases and RTD using a Web site that
operates according to standard Web retailing design principles, the
ID can be presented after the purchase is paid for by means of a
confirmation Web page or in an automatically-sent email. However,
since consumers typically want immediate gratification, and the
finalization event for a particular RTD might not be allowed for
some time, many embodiments will not include features for consumers
to purchase RTD's.)
[0519] In some simple embodiments, the original owner simply sells
RT's to the scouts at a fixed low price. In such cases, a fixed
number of RT's are usually made available for this purpose; later
RT's are made available to consumers without first being made
available to scouts.
[0520] Records representing the status of each RT are stored in a
database (which may be a RAM-based data structure, a disk-based
structure, or a structure in another storage medium). In various
embodiments, there may be one record per RT, or RT's may be
represented in blocks. In most block-based embodiments, a record
will represent a block of RTD's purchased at one time by one scout.
Other representations are equally workable and are equivalently
included in the present invention.
[0521] In most embodiments, the database provides an indicator of
availability of an RT. An RT is available if its current owner is
willing to sell it. In some embodiments, the state of
unavailability is indicated simply by deleting the RT's record from
the database. In others, there is a flag indicating availability or
unavailability.
[0522] In most embodiments, an RTD is automatically made
unavailable when a download occurs. (Equivalently, in embodiments
involving RTE's, the RTE is made unavailable when the experience
occurs; in RTB's it happens when the object is bought.) In some
cases, such as some embodiments involving RTB's, an object may be
made available again at a later date by switching the flag. This is
not the case for RT's which by virtue of their nature may only
occur once.
[0523] In some embodiments, the original owner may choose, at any
time, whether the next sequence of RT's is to be made available to
scouts or to consumers. In some embodiments this is accomplished by
means of a "resellable" indicator in the database. If reselling is
not allowed for a particular RT, then it is of no use to
speculators and they won't want to buy it.
[0524] In some embodiments, RT's for particular sequence numbers
are not entered into the database until a commitment has been made
to sell them.
[0525] However, preferred embodiments perform this function by
means of "minimum price" data in the database, which may be stored
with a separate record for each RT or for blocks of RT's (or as an
indicator that applies to all future RT's, at least until the
indicator is changed). If the minimum price is the maximum price
the consumers are likely to pay, then the RT will be of no use to
scouts.
[0526] A central server (or set of servers working in concert)
keeps track of finalization events.
[0527] The database contains information regarding the price for
finalization events. In some embodiments, scouts and retailers can
set the finalization price for RT's they have purchased. In others,
the finalization price is fixed at the outset by the original
owner. In preferred embodiments, scouts and retailers can set the
price so long as it is under a maximum price fixed by the original
owner, which may have a system-wide default if the original owner
does not specify such a price. This prevents one hostile scout or
retailer from halting sales by setting the finalization price of a
RT so high that no consumer will buy it; since the RT's are
finalized sequentially, lower-priced, subsequent RT's would then
never be sold.
[0528] In preferred embodiments RT's may be purchased
out-of-sequence; that is, for example, a particular scout may
believe that a song will sell 100,000 copies while most scouts
think it will sell 50,000. Therefore in an auction setting, the
scout or retailer may be able to buy the RT's associated with the
90,000th through 100,000th finalizations at a bargain price
compared to the earlier finalizations; if he is right, he may make
an exceptional profit. A user interface is provided whereby the
scout or retailer can specify the range of finalization numbers he
wants to buy at a certain price; in most embodiments other scouts
or retailers may be given the chance to outbid him in an auction;
i.e., it is made known via the user interface that someone has bid
on a particular range of finalizations and the opportunity is
presented to input counter-bids. Any standard auction mechanism
such as Dutch auctions may be used for this.
[0529] In some embodiments RT's don't have to be purchased in
sequential blocks according to finalization sequence; for instance,
scouts and retailers can purchase every nth finalization between
two numbers. This enables them to invest in a wider range of
finalizations depending on how confident they are in the number of
RT's they expect to be sold without buying a huge number of
them.
[0530] In this preferred embodiment, song finalizations, that is,
the actual downloads, are for a fixed price. That way, the value of
the nth RT is simply dependent on the perceived probability that n
or more of the RT's will be finalized. The owner of an RT cannot
refuse a sale; when n-1th finalizations have been sold, the nth one
will be sold next.
[0531] The preferred embodiment draws a clear distinction between
RT's and finalizations (which may be a download or a purchase of a
physical CD, or take other forms).
[0532] There is a speculator market for RT's.
[0533] In the preferred embodiment, there is not a speculator
market for finalizations. There is no need for the price of
finalizations to be variable. For consumer-friendliness--that is,
for the sake of consumers who just want their music and don't want
to hear about speculation--and for retailers that want to keep
everything as simple as possible--finalizations are sold as the
associated objects always have been. Finalizations for CD-related
RT's for instance, are sold on the same fixed-price basis under
which CD's can be purchased on Amazon.com In fact, they may be sold
through Amazon.com.
[0534] These prices will not be out of line with the norms for CD
prices.
[0535] Since the normal fixed price for a finalization means that
it would be absurd for RT's to sell for more than that price, that
creates a natural limit for the price of RT's.
[0536] In this preferred embodiment, original owners can put any
number of RT's on the speculator market at any time. They go into
an auction, and will therefore receive the highest price any
speculator is willing to pay for each RT. Further trades of RT's
take place in a market setting.
[0537] It is to be expected that RT's associated with finalizations
that are far in the future will sell for less than RT's associated
with immediately upcoming ones.
[0538] For example, say 100,000 copies of a CD have been sold, and
the original owner now decides to sell more RT's to the
speculators. It is extremely likely that at least 100,001 CD's will
be sold, so the price for the next RT is likely to be very close to
the regular price for the CD. However, if the speculator decides to
sell 900,000 RT's, then the "IPO" price for the 1,000,000 one might
be very, very low, because it may not be at all obvious that the CD
will ever sell 1,000,000 copies. The speculator's skill--the area
in which they make their profits--is in judging how many copies a
particular work will sell. Our sequential approach, described in my
most recent patent application, greatly enhances the ability of a
speculator to profit from that skill; it removes many factors that
could distort his profits.
[0539] Since RT's are finalized in sequence, if a consumer buys an
RT from the speculators market, he may not be able to finalize it
for some time.
[0540] But there is a queue where consumers can buy the next RT to
be finalized, separate from the speculator's market--actually they
are not really buying an RT at all, they are buying a finalization,
which will be immediate because they are buying the next
finalization.
EXAMPLE 1
[0541] An original owner wants to make 100,000 RT's available for
sale for a particular item, for instance, a recording of a
particular song that will be downloaded. The price will be the same
for every download, $0.25. He wants to sell 10,000 downloads
directly, but since he isn't sure that the song will sell more
downloads than that, he makes subsequent download available to the
speculator's market. See FIG. 2. The original owner causes 90
records to be entered into database table RTTable (1), each of
which represents a sales unit of 1,000 RT's. Status for every
record is set to Avail (meaning that the RT is available to be
purchased). SpeculatorID is set to null. ItemID is an identifier of
the particular song that will be downloaded. (We will assume it is
104). That is, this table may contain RT information for many
different songs; the ItemID allows us to associate a particular
record with a particular song. Each record has a unique (within the
ItemID) SeqNo in the range of 11 to 100. That is, SeqNo's may not
be unique in the overall table, but combined with the ItemID
comprise a unique key into the table. They start at 11 in this case
because the original owner has already committed the first 10,000
downloads to be sold by him directly. FinalizedCount is set to 0
because none of the RT's represented by this block have been
finalized yet. See FIG. 3.
[0542] A set of auctions is arranged, through methods similar to
those on eBay, whereby these blocks of 1,000 RT's are sold. In
fact, the auctions could be conducted on eBay.
[0543] Now, at some point soon after the song has been released a
speculator, represented in SpeculatorTable 2 by the record with
SpeculatorID=452, comes to believe that the song is going to sell
100,000 copies. He places a bid on the block represented by the
record with ItemID 104 and SeqNo 100.
[0544] Assume he wins the auction with a bid of $100. Now Status is
set to Purchased, and RTTable.SpeculatorID is set to 452. See FIG.
4.
[0545] Before and after this purchase, other speculators will have
been purchasing other blocks.
[0546] Separately, downloads are available to consumers. The first
10,000 have no effect on our database because they were not in the
speculator's market. Subsequent downloads update our database.
Continuing with our example of speculator 452, assume that the
90,001st download occurs. The RTTable record with ItemID 104 and
SeqNo 100 is retrieved and FinalizedCount is changed to 1. The
record is saved back into the database.
[0547] This change to the database represents the fact that one
download has been conducted against speculator 452's block. Each
time a download occurs after that, FinalizedCount is incremented
for the same record until it reaches 1,000. (By that point the
original owner may have entered some more RTTable records to
represent succeeding downloads, or he may sell succeeding ones
directly).
[0548] When FinalizedCount reaches 1,000, the SpeculatorTable
record with SpeculatorID 452 is retrieved and payment for 1,000
downloads, is made into his bank account. This payment is $250
minus some processing fees, which in our example happen to be 5%.
So $237.50 is deposited into the bank account, and he has made a
net profit of $137.50 for correctly identifying that 100,000
downloads would occur. See FIG. 5.
[0549] Note that Example 1 is only to be considered as an example.
To list a few of the many variations that could occur: Different
numbers of downloads can be represented by a record other than
1,000, including, for example, 1. Instead of representing
downloads, the records may represent listens to a song (RTE's), or
RTB's. Instead of purchasing the RT's with money, the RT's could be
purchased with some other valuable such as points earned by doing a
useful service. (For example, on the Emergent Music web site,
http://www.emergentmusic.com, points are earned by accurately
rating music and by recommending music that others subsequently
find to be worthwhile.) The subsequent payment, however, could be
in money, or in points that have some other kind of value. As
another variation, the original owner could share in the profits of
each finalization; that is the consumer may pay $0.25 for a
download, where $0.0125 went to transaction fees and $0.10 went to
the original owner, leaving $0.1375 for the speculator. Many other
variations are possible.
[0550] FIG. 1 shows the operation of an exemplary embodiment of the
present invention. In order to establish a market, first a
financial right is determined 13. The financial right is
established as a right which can be sold. This becomes a right
which is the subject of the transaction, referenced as RT 15. The
RT can be a right to download (RTD) 21, a right to experience (RTE)
22 or a right-to-buy (RTB) 23.
[0551] A finalization event is defined and the finalization event
takes place 25. The finalization event may be one or more of object
downloaded 31 in the case of an RTD, object is experienced 32 in
the case of an RTE or physical object is bought 33 in the case of
an RTB. The finalization event is deemed a purchase 35 of the
RT.
[0552] The finalization events are permitted to occur in a
sequential order. That order is established as a sequence, so that
the RT's are finalized in the sequence 37. The RT's are then sold
39, and records representing status of each RT are stored 41 in a
database.
[0553] FIG. 2 is a diagram showing database tables in accordance
with one embodiment of the present invention. An RT table 61 and a
speculator table 62 are shown. The RT table 61 includes an item ID,
a sequence numbers, a status indication, a speculator ID, and a
count of finalized transactions. The speculator table 62 includes a
speculator ID which should correspond to the speculator ID of the
RT table. The speculator table 62 also includes a name, address and
bank account for the speculator identified in the speculator
ID.
[0554] FIG. 3 is a flowchart showing original owner creating RT's
in accordance with one embodiment of the present invention. As can
be seen, the original owner decides to make a particular number of
RT's available. These appear as the item ID in the RT table 61. The
RT's are provided in blocks as desired by the original owner. The
corresponding rows are added to the RT table with sequence numbers.
The items in each row are given the appropriate values in the RT
table.
[0555] FIG. 4 is a flowchart showing a speculator buying RT's in
accordance with one embodiment of the present invention. The
speculator decides to buy a particular block of RT's. The record is
then updated in the RT table.
[0556] FIG. 5 is a flowchart showing a consumer finalizing an RT in
accordance with one embodiment of the present invention. The
consumer causes the finalization of the RT. The RT table is then
updated for that particular sequence number.
* * * * *
References