U.S. patent application number 16/859346 was filed with the patent office on 2020-08-13 for online serving threshold and delivery policy adjustment.
The applicant listed for this patent is Twitter, Inc.. Invention is credited to Qing Zhang.
Application Number | 20200258096 16/859346 |
Document ID | 20200258096 / US20200258096 |
Family ID | 1000004784674 |
Filed Date | 2020-08-13 |
Patent Application | download [pdf] |
![](/patent/app/20200258096/US20200258096A1-20200813-D00000.png)
![](/patent/app/20200258096/US20200258096A1-20200813-D00001.png)
![](/patent/app/20200258096/US20200258096A1-20200813-D00002.png)
![](/patent/app/20200258096/US20200258096A1-20200813-D00003.png)
![](/patent/app/20200258096/US20200258096A1-20200813-D00004.png)
![](/patent/app/20200258096/US20200258096A1-20200813-D00005.png)
United States Patent
Application |
20200258096 |
Kind Code |
A1 |
Zhang; Qing |
August 13, 2020 |
ONLINE SERVING THRESHOLD AND DELIVERY POLICY ADJUSTMENT
Abstract
The present invention provides techniques for use in association
with online advertising, relating to use of serving thresholds,
associated with predicted click through rates, and delivery
policies, associated with advertising inventory serving and
distribution. An offline-trained machine learning-based model may
be utilized in advertising serving decision-making in connection
with serving opportunities, However, serving thresholds and
delivery policies, for use in association with the model in serving
decision-making, may he adjusted online, such as in real-time or
near real-time, based on information obtained online affecting
factors such as predicted click through rates and advertising
inventory distribution.
Inventors: |
Zhang; Qing; (Sunnyvale,
CA) |
|
Applicant: |
Name |
City |
State |
Country |
Type |
Twitter, Inc. |
San Francisco |
CA |
US |
|
|
Family ID: |
1000004784674 |
Appl. No.: |
16/859346 |
Filed: |
April 27, 2020 |
Related U.S. Patent Documents
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
|
|
15664716 |
Jul 31, 2017 |
10672011 |
|
|
16859346 |
|
|
|
|
12764732 |
Apr 21, 2010 |
9754266 |
|
|
15664716 |
|
|
|
|
Current U.S.
Class: |
1/1 |
Current CPC
Class: |
G06Q 30/0244 20130101;
G06Q 30/02 20130101; G06Q 30/0242 20130101 |
International
Class: |
G06Q 30/02 20060101
G06Q030/02 |
Claims
1. A method comprising: during an offline period, determining a set
of serving thresholds for online advertisement serving; during an
offline period, determining a set of delivery policies for
distribution of advertising inventory across advertisement serving
opportunities; during an online period, adjusting at least one of
the set of serving thresholds to determine at least one adjusted
serving threshold, and adjusting at least one of the set of
delivery policies to determine at least one adjusted delivery
policy; and during an online period, modifying serving of online
advertisements in response to advertisement serving opportunities
based on a machine-learning model that is trained on advertising
information associated with serving advertisements during an
offline period and updated during an online period, and further
based at least in part on the at least one adjusted serving
threshold and the at least one adjusted delivery policy, adjusted
during the online period.
2. The method of claim 1, wherein the advertising information
includes user behavior information.
3. The method of claim 1, wherein the advertising information
includes advertisement performance information.
4. The method of claim 1, wherein the advertising information
includes advertisement inventory information.
5. The method of claim 1, wherein the advertising information
includes advertisement distribution information.
6. The method of claim 1, wherein the advertising information is
associated with a node of a hierarchical behavioral targeting
taxonomy.
7. The method of claim 1, further comprising periodically training
the machine-learning model with the advertising information.
8. The method of claim 1, further comprising adjusting at least one
of the set of serving thresholds and at least one of the set of
delivery policies based on a news event occurring during an online
period.
9. The method of claim 1, further comprising adjusting at least one
of the set of serving thresholds and at least one of the set of
delivery policies based on a chronological event occurring during
an online period.
10. The method of claim 1, further comprising: during an offline
period, training the machine-learning model with the advertising
information; and during an online period, adjusting at least one of
the set of serving thresholds and at least one of the set of
delivery policies based on a news event or a chronological event
occurring during an online period and prior to re-training the
machine-learning model during a next offline period based on
advertising information generated during the online period and
associated with the new event or with the chronological event.
11. A system comprising one or more computers and one or more
storage devices on which are stored instructions that are operable,
when executed by the one or more computers, to cause the one or
more computers to perform operations comprising: during an offline
period, determining a set of serving thresholds for online
advertisement serving; during an offline period, determining a set
of delivery policies for distribution of advertising inventory
across advertisement serving opportunities; during an online
period, adjusting at least one of the set of serving thresholds to
determine at least one adjusted serving threshold, and adjusting at
least one of the set of delivery policies to determine at least one
adjusted delivery policy; and during an online period, modifying,
based on a machine-learning model that is trained on advertising
information associated with serving advertisements during an
offline period and updated during an online period, serving of
online advertisements in response to serving opportunities based at
least in part on the at least one adjusted serving threshold and
the at least one adjusted delivery policy, adjusted during the
online period.
12. The system of claim 11, the operations further comprising:
during an offline period, training the machine-learning model with
the advertising information; and during an online period, adjusting
at least one of the set of serving thresholds and at least one of
the set of delivery policies based on a news event or a
chronological event occurring during an online period and prior to
re-training the machine-learning model during a next offline period
based on advertising information generated during the online period
and associated with the new event or with the chronological
event.
13. The system of claim 11, wherein the advertising information
includes user behavior information.
14. The system of claim 11, wherein the advertising information
includes advertisement performance information.
15. The system of claim 11, wherein the advertising information
includes advertisement inventory information.
16. The system of claim 11, wherein the advertising information
includes advertisement distribution information.
17. The system of claim 11, wherein the advertising information is
associated with a node of a hierarchical behavioral targeting
taxonomy.
18. The system of claim 11, the operations further comprising
adjusting at least one of the set of serving thresholds and at
least one of the set of delivery policies based on a news event or
a chronological event occurring during an online period.
19. A non-transitory computer readable medium or media containing
instructions for executing a method for use in association with an
online advertising marketplace, the method comprising: during an
offline period, determining a set of serving thresholds for online
advertisement serving; during an offline period, determining a set
of delivery policies for distribution of advertising inventory
across advertisement serving opportunities; during an online
period, adjusting at least one of the set of serving thresholds to
determine at least one adjusted serving threshold, and adjusting at
least one of the set of delivery policies to determine at least one
adjusted delivery policy; and during an online period, modifying
serving of online advertisements in response to advertisement
serving opportunities based on a machine-learning model that is
trained on advertising information associated with serving
advertisements during an offline period and updated during an
online period, and further based at least in part on the at least
one adjusted serving threshold and the at least one adjusted
delivery policy, adjusted during the online period.
20. The non-transitory computer readable medium of claim 19, the
method further comprising: during an offline period, training the
machine-learning model with the advertising information; and during
an online period, adjusting at least one of the set of serving
thresholds and at least one of the set of delivery policies based
on a news event or a chronological event occurring during an online
period and prior to re-training the machine-learning model during a
next offline period based on advertising information generated
during the online period and associated with the new event or with
the chronological event.
21. The non-transitory computer readable medium of claim 19, the
operations further comprising adjusting at least one of the set of
serving thresholds and at least one of the set of delivery policies
based on a news event or a chronological event occurring during an
online period.
Description
CROSS-REFERENCE TO RELATED APPLICATIONS
[0001] This application is a continuation of U.S. application Ser.
No. 15/664,716, filed Jul. 31, 2017, which is a continuation of
U.S. application Ser. No. 12/764,732, filed Apr. 21, 2010, the
entire disclosures of which are incorporated by reference.
BACKGROUND
[0002] In targeting, such as behavioral targeting, historical
information such as online user behavior information can be used in
targeting high-performing advertisements to users. In this regard,
a taxonomy, such as a hierarchical taxonomical tree of categories
or topics, may be utilized, in which nodes in the taxonomy may
represent categories of interest or behavioral targeting
categories. A machine learning-based model may be utilized in
selecting advertisements for matching with serving opportunities
and for serving to particular users. Historical user behavior
information may he used to train the model offline. For example,
such models may be trained periodically, such as monthly, weekly,
or even more frequently, using updated historical user behavior
information.
[0003] Online, the offline-trained model may be utilized in
advertisement targeting and in determining or facilitating
determination of such things as serving thresholds and delivery
policies. Serving thresholds may include scores, which may directly
correspond to particular CTRs, for instance. The serving thresholds
may, for instance, relate to certain minimum scores or CTRs that
will be required for serving of an advertisement in a particular
category of the taxonomy, for instance (of course, there are many
details in the process which are not described here, for simplicity
of explanation). Delivery policies may, for instance, govern
serving based at least in part on available advertising inventory,
or available inventory for a. particular taxonomical category, etc.
For example, serving thresholds may be set in each category in such
a way that both a desired level of performance, measured by
category-specific CFR, and a desired volume of deliverable ad
impressions in the same category, are predicted to be achieved.
Naturally, for a large-scale operation, scores and thresholds may
be set based on many other factors as well, and will take into
account many other variables across many advertisers, campaigns,
etc. Generally, the model may be used in making predictions and
projections based on the offline training.
[0004] Generally, serving thresholds and delivery policies are
determined offline. Online, real-time or near real-time
information, including newly obtained user behavior information,
etc., can be fed into the model, and the model can be used in
determining when circumstances are right for serving, such as when
a particular serving opportunity to a particular user is predicted
or projected by the model to meet requirements such as the
predetermined serving thresholds and delivery policy
requirements.
[0005] As mentioned, such models are generally refreshed
periodically by offline training with newly collected historical
user behavior information. Models may only be refreshed so
frequently, such as monthly, weekly, or perhaps daily. However,
circumstances, events, and developments occur and change
dynamically in real time, and the model cannot be refreshed
constantly to include such information as training information.
Such dynamic developments may include, as just one example, a
breaking news event, which may affect anticipated CTRs or pertain
to optimal delivery policies, etc. Existing methods utilizing, for
example, serving thresholds and delivery policies set using models
that may have been refreshed offline may lead to suboptimal
serving-related decision-making and determinations.
[0006] There is a need for methods and systems for improving or
optimizing serving decision-making and determinations and
associated serving thresholds and delivery policies.
SUMMARY
[0007] Some embodiments of the invention provide methods and
systems for use in association with online advertising, relating to
use of serving thresholds, associated with predicted click through
rates, and delivery policies, associated with advertising inventory
serving and distribution. An offline-trained machine learning-based
model may be utilized in advertising serving decision-making in
connection with serving opportunities. However, serving thresholds
and delivery policies, for use in association with the model in
serving decision-making, may be adjusted online, such as
dynamically in real-time or near real-time, based on information
obtained online affecting factors such as predicted click through
rates and advertising inventory distribution.
[0008] In some embodiments, for example, use of an offline-trained
machine learning- based model in combination with online-adjusted
serving thresholds and delivery policies can provide for efficient
and effective techniques for optimizing, or better optimizing,
behavioral targeting and advertising inventory distribution.
BRIEF DESCRIPTION OF THE DRAWINGS
[0009] FIG, 1 is a distributed computer system according to one
embodiment of the invention;
[0010] FIG. 2 is a flow diagram illustrating a method according to
one embodiment of the invention;
[0011] FIG. 3 is a flow diagram illustrating a method according to
one embodiment of the invention;
[0012] FIG. 4 is a flow diagram illustrating a method according to
one embodiment of the invention; and
[0013] FIG. 5 is a flow diagram illustrating a method according to
one embodiment of the invention.
[0014] While the invention is described with reference to the above
drawings, the drawings are intended to be illustrative, and the
invention contemplates other embodiments within the spirit of the
invention,
DETAILED DESCRIPTION
[0015] FIG. 1 is a distributed computer system 100 according to one
embodiment of the invention. The system 100 includes user computers
104, advertiser computers 106 and server computers 108, all coupled
or able to be coupled to the Internet 102. Although the Internet
102 is depicted, the invention contemplates other embodiments in
which the Internet is not included, as well as embodiments in which
other networks are included in addition to the Internet, including
one more wireless networks, WANs, LANs, telephone, cell phone, or
other data networks, etc. The invention further contemplates
embodiments in which user computers or other computers may be or
include wireless, portable, or handheld devices such as cell
phones, PDAs, etc.
[0016] Each of the one or more computers 104, 106, 108 may be
distributed, and can include various hardware, software,
applications, algorithms, programs and tools. Depicted computers
may also include a hard drive, monitor, keyboard, pointing or
selecting device, etc. The computers may operate using an operating
system such as Windows by Microsoft, etc. Each computer may include
a central processing unit (CPU), data storage device, and various
amounts of memory including RAM and ROM, Depicted computers may
also include various programming, applications, algorithms and
software to enable searching, search results, and advertising, such
as graphical or banner advertising as well as keyword searching and
advertising in a sponsored search context. Many types of
advertisements are contemplated, including textual advertisements,
rich advertisements, video advertisements, etc.
[0017] As depicted, each of the server computers 108 includes one
or more CPUs 110 and a data storage device 112. The data storage
device 112 includes a database 116 and a Serving Threshold and
Delivery Policy Adjustment Program 114.
[0018] The Program 114 is intended to broadly include all
programming, applications, algorithms, software and other and tools
necessary to implement or facilitate methods and systems according
to embodiments of the invention. The elements of the Program 114
may exist on a single server computer or be distributed among
multiple computers or devices.
[0019] FIG. 2 is a flow diagram illustrating a method 200 according
to one embodiment of the invention. At step 202, using one or more
computers, during an offline period, a set of serving thresholds is
initially determined, to be utilized in online advertisement
serving, in which a serving threshold is associated with a minimum
anticipated click through rate.
[0020] At step 204, using one or more computer computers, during an
offline period, a set of delivery policies is initially determined,
in which a delivery policy is associated with one or more rules
relating to serving of advertisements in accordance with required
or optimal distribution of advertising inventory across serving
opportunities.
[0021] At step 206, using one or more computers, during, and based
at least in part on information obtained during, an online period,
adjustment is performed of at least one of the set of serving
thresholds to determine at least one adjusted serving threshold,
and adjustment is performed of at least one of the set of delivery
policies to determine at least one adjusted delivery policy.
[0022] At step 208, using one or more computers, during an online
period, a machine learning-based model is utilized in
decision-making with regard to serving of online advertisements in
connection with serving opportunities based at least in part on the
at least one adjusted serving threshold and the at least one
adjusted delivery policy, in which the machine learning-based model
is trained during an offline period, and in which an online period
is a period of active advertisement serving in which the model is
utilized.
[0023] FIG. 3 is a flow diagram illustrating a method 300 according
to one embodiment of the invention, At step 302, using one or more
computers, during an offline period, a set of serving thresholds
are initially determined, to be utilized in online advertisement
serving, in which a serving threshold is associated with a minimum
anticipated click through rate.
[0024] At step 304, using one or more computer computers, during an
offline period, a set of delivery policies is initially determined,
in which a delivery policy is associated with one or more rules
relating to serving of advertisements in accordance with required
or optimal distribution of advertising inventory across serving
opportunities. The distribution of advertising inventory across
serving opportunities relates to distribution of advertising
inventory across serving opportunities associated with different
nodes of a hierarchical taxonomy of behavioral targeting
categorical nodes.
[0025] At step 306, using one or more computers, during, and based
at least in part on information obtained during, an online period,
adjustment is performed of at least one of the set of serving
thresholds to determine at least one adjusted serving threshold,
and adjustment is performed of at least one of the set of delivery
policies to determine at least one adjusted delivery policy. The
adjusted serving threshold and the adjusted delivery policy are
adjusted at least in part to better optimize user targeting or
advertising inventory distribution, based at least in part on the
circumstances occurring during the period.
[0026] At step 308, using one or more computers, during an online
period, a machine learning-based model is utilized in
decision-making with regard to serving of online advertisements in
connection with serving opportunities based at least in part on the
at least one adjusted serving threshold and the at least one
adjusted delivery policy. The machine learning-based model is
trained during an offline period, using information including
historical user behavior information. Using one or more computers,
during an online period, the machine learning-based model is
utilized in decision-making with regard to serving of online
advertisements in connection with serving opportunities based at
least in part on the at least one adjusted serving threshold and
the at least one adjusted delivery policy. An online period is a
period of active advertisement serving in which the model is
utilized.
[0027] FIG. 4 is a flow diagram illustrating a method 400 according
to one embodiment of the invention. A machine learning-based model
is depicted by block 404. The model 404 may be a featured-based
model. Information, including advertising-related information, is
used as training information for the model 404. The information may
be stored in a data store, such as data store 410. The data store
410 can be or include any of various forms or manners of data
storage, and may or may not include one or more databases. The
information may include, for example, historical user behavior
information, click through rate or other advertisement performance
information, advertisement inventory and distribution information,
and other information. The information may be category-specific,
such as by being specific to nodes of a hierarchical behavioral
targeting taxonomy.
[0028] The model 404 is trained offline periodically, as
represented by block 402. For example, the model may be trained
monthly, weekly, daily, or otherwise.
[0029] Serving thresholds and delivery policies are initially
determined offline, as represented by block 406, such as prior to
an online period. An online period may be a period during which
active online advertisement serving takes place, including use of
the model 404. An offline period may be period during which active
advertisement serving using the model is not taking place, or
during which active advertisement serving associated with a
particular active period is not taking place, such as a period
prior to an active serving period, or prior to a particular active
serving period.
[0030] At step 412, online monitoring, tracking, and storing, such
as in the data store 410, is performed of category-specific
advertising information, including CTR and inventory distribution
information,
[0031] At step 414, serving thresholds and delivery policies are
dynamically adjusted online to optimize based on monitored
advertising information. For example, the serving thresholds and
delivery policies may be adjusted in real-time or near real-time,
based on real- time or near real-time advertising information.
[0032] At step 416, online, the model, as well as adjusted serving
resholds and adjusted delivery policies, are utilized in online
advertising serving determinations/decision-making.
[0033] FIG. 5 is a flow diagram illustrating a method 500
illustrating one embodiment of the invention. At step 502, offline
training is performed of a machine learning-based model.
[0034] At step 504, offline setting of serving thresholds and
delivery policies is performed.
[0035] At step 506, dynamic, such as real-time or near real-time,
monitoring is performed of advertising-relevant information, such
as category-specific advertisement performance and inventory
distribution information, assuming time remains in the pertinent
online period. Advertising-relevant information can also include
many other types of information that may affect advertisement
performance or optimal delivery policies, such as time-based or
news-based developments that may alter predicted click through rate
for a particular behavioral targeting category, for instance.
[0036] At step 508, the method 500 queries whether adjustments to
serving thresholds or delivery policies are indicated based on the
monitored advertising-relevant information. Various embodiments of
the invention contemplate various ways by which particular
adjustments may be determined, and by which underlying anticipated
changes, such as changes in predicted click through rate for a
particular category, for instance, may be determined or
estimated.
[0037] If adjustments to serving thresholds and delivery policies
are not called for based on presently available monitored
information, then the method 500 proceeds to step 512.
[0038] If adjustments are called for, then corresponding
adjustments are made at step 510, and then the method 500 proceeds
to step 512.
[0039] At step 512, thresholds (including any currently adjusted
thresholds) and the model 404 are utilized in targeting and serving
determinations.
[0040] Following step 512, the method 500 return to 506, from which
point newly available information can be monitored and utilized, if
time remains in the pertinent online period.
[0041] Generally, some embodiments of the invention provide a
framework that can be utilized to adjust serving thresholds and
delivery policies, in online advertising, such as to improve
targeting performance or inventory distribution based on
dynamically, real-time or near real-time monitored category
specific (such as behavioral targeting category-specific) click
through rate (or other advertisement performance) or inventory
information.
[0042] As mentioned, some embodiments of the invention include use
of an offline- trained machine learning-based model, which makes
use of information including historical user information, in
determinations and decision-making regarding serving of particular
advertisements to particular users and in connection with
particular serving opportunities. As a simple example, the model
may be used online in dynamically updating a score for a particular
user based on circumstances that may include activity of the user.
The score may correspond with a predicted click through rate for
the user in connection with a particular behavioral targeting
category, for instance. If a user then appears online, for example,
at a time within a particular targeting time-frame, and the user's
score at that time is at or above a particular threshold, then an
advertisement corresponding to that category may be served to the
user in connection with an associated serving opportunity. Serving
thresholds and delivery policies may he utilized, for example, in
achieving desired levels of advertisement performance in connection
with particular categories of a hierarchical taxonomy of user
interest categories.
[0043] In some embodiments of the invention, thresholds and
policies, initially determined offline, are adjusted online in
accordance with monitored online information, For instance,
predicted click through rate may vary with many real-time or
online-occurring circumstances, such as time of time of day or
week, or the occurrence of a breaking news or other sudden or
real-time occurring or developing event, etc. The model, trained
periodically offline, has not been refreshed to take into account
this information. However, adjustment of serving thresholds and
serving policies can be utilized to account for and optimize based
on the monitored online information. The model can then be utilized
along with the adjusted thresholds and serving policies to produce
better optimized results. For example, the thresholds and policies
can be adjusted in order to, for example, better achieve desired
levels of performance in connection with particular categories,
better distribution of advertising inventory across categories,
etc.
[0044] For instance, a sudden event such as the death of an actor,
or a celebrity scandal, can trigger a time-dependent spike in user
interest in certain categories. As another example, shopping
activity may be monitored to peak during certain periods or
holidays, or as a result of current events, such as prior to a
predicted. rain storm or snow storm, for instance. Adjusted
thresholds and policies can be used to better optimize, effectively
and efficiently allowing taking into account anticipated effects of
such circumstances or events, which anticipated effects and the
degree of such effects may be estimated or calculated in any of
various ways.
[0045] In some embodiments, methods are utilized that include
online monitoring of inventory or inventory distribution, and CTR,
from the specified categories, such as from or in a real-time or
near real-time online behavioral targeting scoring system (in which
scores may correspond directly with predicted CTRs, for instance),
and an advertisement serving system.
[0046] For instance, inventory information can be derived from or
in part from any of various metrics. These may include, for
example, type and number of events affecting a specified category
during given time window, number of unique users affected in a
specified category during a given time window, number of affected
users whose interest scores are above current category-specific
serving threshold, for example, and various other metrics and
information. In some embodiments, a real-time or near real-time
behavioral targeting system may obtain all advertising click and
view event information, allowing efficient calculation of
category-specific CTR in a given time frame, for instance. Once the
category-specific CTR is obtained, it can be utilized in various
ways and for various purposes, including calculating and
determining serving threshold adjustments, delivery policy
adjustments, etc., in real-time or near real-time.
[0047] In some embodiments, serving policies may be adjusted, for
example, to prioritize serving in particular categories. As a
simple example, a particular user may be qualified to receive
advertisements from a number of categories, but available
applicable serving opportunities to the user may mean that only
advertisements from some of the categories may be served. In such
circumstances, if monitored online information leads to a
determination that predicted CTR for a particular category is
currently above that which was projected using the offline-trained
model, then that category can be prioritized in terms of
advertisement serving, such as by adjusting a delivery policy to
emphasize or prioritize advertisement serving in that category.
[0048] In some embodiments, for example, a ratio of the current
predicted CTR for a particular category to the projected CTR for
the particular category using the offline-trained model, for a
particular time window, can be used as at least one factor in
determining delivery policy, where, for instance, a ratio greater
than I may lead to a delivery policy adjustment that prioritizes
that category for that time window. Furthermore, lower online-
predicted CTRs can lead to adjustments that de-emphasize associated
categories, etc. Over time, by using this technique repeatedly over
time, a desired inventory distribution can be maintained while
still increasing overall CTR, leading to increased revenue, etc.
Generally, in some embodiments, online traffic patterns and pattern
changes can be monitored and leveraged, through adjustments to
serving thresholds and delivery policies, to better achieve or
maintain online advertising goals and metrics, for instance, in an
efficient and cost- effective manner.
[0049] While the invention is described with reference to the above
drawings, the drawings are intended. to be illustrative, and the
invention contemplates other embodiments within the spirit of the
invention.
* * * * *