U.S. patent application number 13/167591 was filed with the patent office on 2011-10-20 for autocompletion of partial search query with return of predicted search results.
Invention is credited to Kevin A. Gibbs, Taher H. Haveliwala, Glen M. Jeh, Sepandar D. Kamvar.
Application Number | 20110258183 13/167591 |
Document ID | / |
Family ID | 40298150 |
Filed Date | 2011-10-20 |
United States Patent
Application |
20110258183 |
Kind Code |
A1 |
Gibbs; Kevin A. ; et
al. |
October 20, 2011 |
Autocompletion of Partial Search Query with Return of Predicted
Search Results
Abstract
A set of ordered predicted completion strings are presented to a
user as the user enters text in a text entry box (e.g., a browser
or a toolbar). The predicted completion strings can be in the form
of URLs or query strings. The ordering may be based on any number
of factors (e.g., a query's frequency of submission from a
community of users). URLs can be ranked based on an importance
value of the URL. Privacy is taken into account in a number of
ways, such as using a previously submitted query only when more
than a certain number of unique requestors have made the query. The
sets of ordered predicted completion strings is obtained by
matching a fingerprint value of the user's entry string to a
fingerprint to table map which contains the set of ordered
predicted completion strings.
Inventors: |
Gibbs; Kevin A.; (San
Francisco, CA) ; Kamvar; Sepandar D.; (Palo Alto,
CA) ; Haveliwala; Taher H.; (Mountain View, CA)
; Jeh; Glen M.; (San Francisco, CA) |
Family ID: |
40298150 |
Appl. No.: |
13/167591 |
Filed: |
June 23, 2011 |
Related U.S. Patent Documents
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
|
|
12345564 |
Dec 29, 2008 |
|
|
|
13167591 |
|
|
|
|
10987295 |
Nov 11, 2004 |
7487145 |
|
|
12345564 |
|
|
|
|
10875143 |
Jun 22, 2004 |
7836044 |
|
|
10987295 |
|
|
|
|
Current U.S.
Class: |
707/723 ;
707/E17.014 |
Current CPC
Class: |
G06F 16/3322 20190101;
Y10S 707/99934 20130101; Y10S 707/99935 20130101; Y10S 707/99943
20130101 |
Class at
Publication: |
707/723 ;
707/E17.014 |
International
Class: |
G06F 17/30 20060101
G06F017/30 |
Claims
1. A computer-implemented method for processing query information,
comprising: at a server system: receiving a partial search query
from a client system, the receiving including receiving the partial
search query from the client system prior to receiving from the
client system a search request on a complete search query that
includes the partial search query; responding to receiving the
partial search query by: obtaining a set of historical complete
search queries, the set of historical complete search queries
corresponding to the partial search query and ordered in accordance
with a ranking criteria; obtaining predicted search results
corresponding to one or more search queries of the set of
historical complete search queries; and conveying the predicted
search results to the client system.
2. The method of claim 1, wherein the predicted search results are
conveyed to the client system prior to receiving from the client
system the search request on the complete search query that
includes the partial search query.
3. The method of claim 1, wherein the predicted search results
correspond to a predicted search query associated with subjects,
concepts, or categories of information that are of interest to the
user.
4. The method of claim 1, wherein the predicted search results
correspond to a predicted search query previously submitted by the
user.
5. The method of claim 1, wherein the predicted search results are
formatted for display in a portion of a non-persistent window.
6. The method of claim 1, wherein the predicted search results
include snippets of content from one or more documents identified
by the predicted search results.
7. The method of claim 1, wherein the predicted search results
include thumbnails of content from one or more documents identified
by the one or more predicted search results.
8. The method of claim 1, wherein the predicted search results
include thumbnails of one or more web pages identified by the one
or more predicted search results.
9. The method of claim 1, further comprising: conveying to the
client system along with the predicted search results, the set of
historical complete search queries.
10. The method of claim 9, wherein the predicted search results and
the set of historical complete search queries are formatted for
display in single display window.
11. A search engine system, for processing query information,
comprising: one or more processors; and memory storing one or more
programs to be executed by the one or more processors; the one or
more programs comprising instructions for: receiving a partial
search query from a client system, the receiving including
receiving the partial search query from the client system prior to
receiving from the client system a search request on a complete
search query that includes the partial search query; responding to
receiving the partial search query by: obtaining a set of
historical complete search queries, the set of historical complete
search queries corresponding to the partial search query and
ordered in accordance with a ranking criteria; obtaining predicted
search results corresponding to one or more search queries of the
set of historical complete search queries; and conveying the
predicted search results to the client system.
12. The search engine system of claim 11, further comprising
instructions for: conveying the predicted search results to the
client system prior to receiving from the client system the search
request on the complete search query that includes the partial
search query.
13. The search engine system of claim 11, wherein the predicted
search results correspond to a predicted search query associated
with subjects, concepts, or categories of information that are of
interest to the user.
14. The search engine system of claim 11, wherein the predicted
search results correspond to a predicted search query previously
submitted by the user.
15. The search engine system of claim 11, further comprising
instructions for: formatting the predicted search results for
display in a portion of a non-persistent window.
16. The search engine system of claim 11, wherein the predicted
search results include snippets of content from one or more
documents identified by the predicted search results.
17. The search engine system of claim 11, further comprising
instructions for: conveying to the client system along with the
predicted search results, the set of historical complete search
queries.
18. The search engine system of claim 11, further comprising
instructions for: formatting the predicted search results and the
set of historical complete search queries for display in single
display window.
19. A non-transitory computer readable storage medium storing one
or more programs configured for execution by a computer, the one or
more programs comprising instructions for: receiving a partial
search query from a client system, the receiving including
receiving the partial search query from the client system prior to
receiving from the client system a search request on a complete
search query that includes the partial search query; responding to
receiving the partial search query by: obtaining a set of
historical complete search queries, the set of historical complete
search queries corresponding to the partial search query and
ordered in accordance with a ranking criteria; obtaining predicted
search results corresponding to one or more search queries of the
set of historical complete search queries; and conveying the
predicted search results to the client system.
20. The search engine system of claim 11, further comprising
instructions for: conveying the predicted search results to the
client system prior to receiving from the client system the search
request on the complete search query that includes the partial
search query.
21. The search engine system of claim 11, wherein the predicted
search results correspond to a predicted search query associated
with subjects, concepts, or categories of information that are of
interest to the user.
22. The search engine system of claim 11, wherein the predicted
search results correspond to a predicted search query previously
submitted by the user.
23. The search engine system of claim 11, further comprising
instructions for: formatting the predicted search results for
display in a portion of a non-persistent window.
24. The search engine system of claim 11, wherein the predicted
search results include snippets of content from one or more
documents identified by the predicted search results.
25. The search engine system of claim 11, further comprising
instructions for: conveying to the client system along with the
predicted search results, the set of historical complete search
queries.
26. The search engine system of claim 11, further comprising
instructions for: formatting the predicted search results and the
set of historical complete search queries for display in single
display window.
Description
RELATED APPLICATIONS
[0001] This application claims priority to and is a continuation of
U.S. patent application Ser. No. 12/345,564, filed Dec. 29, 2008,
entitled "Method and System for Autocompletion Using Ranked
Results," which is a continuation of U.S. patent application Ser.
No. 10/987,295, filed Nov. 11, 2004, now U.S. Pat. No. 7,487,145,
titled "Method and System for Autocompletion Using Ranked Results,"
and is also a continuation-in-part of U.S. patent application Ser.
No. 10/875,143, filed Jun. 22, 2004, now U.S. Pat. No. 7,836,044,
titled "Anticipated Query Generation and Processing in a Search
Engine," which applications are incorporated by reference herein in
their entirety.
TECHNICAL FIELD
[0002] The present invention relates generally to the field of
search engines for locating documents in a computer network (e.g.,
a distributed system of computer systems), and in particular, to a
system and method for speeding up a desired search by anticipating
a user's request.
BACKGROUND
[0003] Search engines provide a powerful tool for locating
documents in a large database of documents, such as the documents
on the World Wide Web (WWW) or the documents stored on the
computers of an Intranet. The documents are located in response to
a search query submitted by a user. A search query may consist of
one or more search terms.
[0004] In one approach to entering queries, the user enters the
query by adding successive search terms until all search terms are
entered. Once the user signals that all of the search terms of the
query have been entered, the query is sent to the search engine.
The user may have alternative ways of signaling completion of the
query by, for example, entering a return character, by pressing the
enter key on a keyboard or by clicking on a "search" button on a
graphical user interface. Once the query is received by the search
engine, it processes the search query, searches for documents
responsive to the search query, and returns a list of documents to
the user.
[0005] Because the query is not sent to the search engine until the
user has signaled that the query is complete, time passes while the
user is finishing the full search query. It would be desirable to
have a system and method of speeding up this process.
SUMMARY
[0006] In one embodiment, a system and method for processing query
information includes receiving query information from a search
requestor, prior to the user indicating completion of the query.
From the received query information a set of predicted queries
ordered in accordance with a ranking criteria is predicted. The set
of ordered predicted queries is then transmitted to the search
requestor.
[0007] The search requestor may select a respective query from the
ordered set of predicted queries and then indicate completion of
the query. A search engine processes the query to produce a set of
search results. Alternately, the search requestor may continue
entering query information until a complete query is entered, or
until a new set of predicted queries is transmitted and presented
to the search requestor.
BRIEF DESCRIPTION OF THE DRAWINGS
[0008] The aforementioned embodiment of the invention as well as
additional embodiments will be more clearly understood as a result
of the following detailed description of the various aspects of the
invention when taken in conjunction with the drawings. Like
reference numerals refer to corresponding parts throughout the
several views of the drawings.
[0009] FIG. 1 depicts a process for predicting queries in
accordance with some embodiments of the present invention.
[0010] FIG. 2 depicts a block diagram of a search system in
accordance with some embodiments of the present invention.
[0011] FIG. 3 depicts a process in a search assistant in accordance
with some embodiments of the present invention.
[0012] FIG. 4 depicts a process for receiving query input and
creating responses thereto in accordance with some embodiments of
the present invention.
[0013] FIG. 5 depicts flows of information associated with creating
and using a fingerprint-to-table map in accordance with some
embodiments of the present invention.
[0014] FIG. 6 depicts examples of relevancy of input strings in
accordance with some embodiments of the present invention.
[0015] FIG. 7 depicts a process for processing historical queries
in accordance with some embodiments of the present invention.
[0016] FIG. 8 depicts a portion of an exemplary table used in
processing historical queries in accordance with some embodiments
of the present invention.
[0017] FIG. 9 depicts data structures associated with a query
completion table using suffixes in accordance with some embodiments
of the present invention.
[0018] FIG. 10 depicts a portion of an exemplary query completion
table in accordance with some embodiments of the present
invention.
[0019] FIG. 11 depicts an exemplary screen shot in accordance with
some embodiments of the present invention.
[0020] FIG. 12 depicts a search engine suitable for implementing
some embodiments of the present invention.
[0021] FIG. 13 depicts a client suitable for implementing some
embodiments of the present invention.
DESCRIPTION OF EMBODIMENTS
[0022] In one embodiment of the invention, portions of a user's
query are transmitted to a search engine before the user has
finished entering the complete query. The search engine uses the
transmitted portion of the query to predict the user's final query.
These predictions are transmitted back to the user. If one of the
predictions is the user's intended query, then the user can select
that predicted query without having to complete entry of the query.
In some embodiments, the selected query is transmitted to the
search engine, which returns a set of query results corresponding
to the selected query.
[0023] FIG. 1 illustrates an exemplary embodiment of the invention
including a client system 104 and a search engine 106. As a user
enters a search query, the user's input is monitored by the client
system (108). Prior to the user signaling completion of the search
query, a portion of the user's query is sent from the client system
104 to the search engine 106 (110). The portion of the query may be
a few characters, a search term, or more than one search term. In
some embodiments, the partial input is in the form of a content
location identifier, often called a uniform resource locator (URL)
such as that described in RFC 1738, promulgated by the Internet
Engineering Task Force, which can be used to identify resources
within computers and computer networks. URLs can also be used to
identify resources available locally on a computer such as
documents, folders or services. The term "URL" is used herein to
mean any form of content location identifier, including but not
limited to Internet addresses, RFC 1738 compliant addresses, and
file pathnames such as those use in many computer systems and local
area networks. The search engine 106 receives the partial query for
processing (112) and makes predictions as to user's contemplated
complete query (or URL) (114). The predictions are ordered
according in accordance with a ranking criteria. For example, in
some embodiments queries having a higher frequency of submission
are ordered before queries having lower frequencies of submission.
The search engine 106 uses a number of query completion tables
(described in more detail below) to assist in making the ordered
predictions. The query completion tables are created using
previously entered search queries received by the search engine
106. In some embodiments, the previous queries include search
queries from a community of users. The predicted queries are sent
back to the client system 106 (116) and then presented to the user
(118). If one of the predicted queries is what the user intended as
the desired query, the user may select this predicted query and
proceed without having to finish entering the desired query. If the
predicted queries do not reflect what the user had in mind, then
the user may continue entering the desired search query.
[0024] FIG. 2 illustrates a searching system 200 according to some
embodiments of the invention and shows various functional
components which will be referred to in the detailed discussion
which follows. The search system 200 may include one or more client
systems 202. Each client system 202 has a search assistant 204. The
client systems 202 are connected to a communications network 206.
The communications network 206 connects the client systems 202 to a
search engine 208. Search engine 208 includes a query server 210
connected to the communications network 206, a prediction server
212 and a query processing controller 214.
[0025] The query server 210 includes a client communications module
216, a query receipt, processing and response module 218, a partial
query receipt, processing and response module 220, a user
information processing module 222, and a query log 224, all
interconnected. In some embodiments, fewer and/or additional
modules or functions are included in the query server 210. The
modules shown in FIG. 2 as being part of query server 210 represent
functions performed in an exemplary embodiment. The prediction
server 212 is connected to partial query receipt, processing and
response module 220, the ordered set builder 242 and to query log
224. The ordered set builder 242 creates sets of ordered predicted
queries from logs of queries and URL requests, and is connected to
the query log 224. In some embodiments, the ordered set builder 242
is also coupled to a URL database 225. In such embodiments, the
prediction server 212 is connected directly to the query log 224
and the URL database 225.
[0026] The query processing controller 214 is connected to an
inverse document index 228, a document database 230, a query cache
232 and the URL database 225. The cache 232 may include an index
234 the function of which is to locate entries in the cached
results 236. The cached results 236 may include a cache entry for
an identified query 238 and a cache entry for an anticipated query
240. The inverse document index 228 and document database 230 are
sometimes collectively called the document database. In some
embodiments, "searching the document database" means searching the
inverse document index 228 to identify documents matching a
specified search query or term.
[0027] Although illustrated as discrete blocks in the figure, FIG.
2 is intended more as a functional description of an embodiment of
the invention rather than a structural mapping of the functional
elements. One of ordinary skill in the art would recognize that an
actual implementation might have the functional elements grouped or
split among various components. For example, the query log 224 may
be distinct from the query server 210. In some embodiments the
query log 224 may be stored on one or more servers whose primary
function is to store and process query log information. Similarly,
the URL database 225 may be stored on or more servers whose primary
purpose is to store and process information about known URLs.
[0028] FIG. 3 illustrates an embodiment of the invention that may
be implemented in the search assistant 204 of a client system 202
(FIG. 2). The search assistant 204 monitors the user's entry of a
search query on the client system 104 (302). In some embodiments,
the search assistant 204 monitors the user's entry of a uniform
resource locator (URL) input string, such as in the address field
of a browser window. The user may enter the search query or URL in
a number of ways including a browser window, a search tool, or any
other input mechanism. The search assistant 204 may identify two
different scenarios. First, the search assistant 204 receives or
identifies a final input (302-final input) when the user has
indicated completion of the input string or selected a presented
prediction. Second, the search assistant 204 receives or identifies
a partial input (302-partial input) when an input is identified
prior to when the user indicates completion of the input string (as
described below). In a third, optional scenario (described in more
detail below), the search assistant 204 determines or receives
notification that the user has not selected one of the predictions
within a specified time period.
[0029] When a final input or selection (302-final input) is
identified as a search query, the input is transmitted to the
search engine 208 (304) for processing. The search engine 208
returns a set of search results, which is received by the search
assistant 204 (306) or by a client application, such as a browser
application. The list of search results is presented to the user
such that the user may select one of the documents for further
examination (e.g., visually or aurally). When the final input is a
URL, the request is transmitted to the appropriate document host
(304) and the document, if available, is returned (306). After the
response is received (306), the user's input activities are again
monitored (302). In some embodiments, the URL request is sent to
the search engine 208 for logging and the request is redirected to
the appropriate document host.
[0030] A final input may be identified by the search assistant 204
in a number of ways such as when the user enters a carriage return,
or equivalent character, selects a search button in a graphical
user interface (GUI) presented to the user during entry of the
search query, or by possibly selecting one of a set of possible
queries presented to the user during entry of the search query. One
of ordinary skill in the art will recognize a number of ways to
signal the final entry of the search query.
[0031] Prior to the user signaling a final input, a partial input
may be identified (302-partial input). A partial input may be
identified in a number of ways. For a search query, a partial input
includes a single search term of the search query, multiple search
terms, or a predefined a number of characters of a search term.
[0032] In some embodiments, a partial input is identified by
detecting entry of delimiter or other character (e.g., without
limitation, a quote character, a period, a parenthesis character, a
slash character, arrow key detection or tab entry). Entry of a
delimiting character may indicate that a user has finished entering
a desired term or portion of the input and is moving onto the next
search term or portion.
[0033] In some embodiments, a partial input is identified by
detecting entry of a pre-determined number of characters. In these
embodiments, the input contains a number of characters less than a
full input but it may still desirable to identify the partial input
before the user has entered all of the characters. This technique
is desirable, for example, when the search term or URL contains a
large number of characters or when the pre-determined number of
characters is large enough to result in useful predictions.
[0034] In some embodiments, a partial input is identified by
detecting the absence of a character being entered within a period
of time, the absence representing a pause by the user. The pause
may signify that the user has entered one search term or portion of
the complete string but has not entered the space key (or other
delimiting character) to start entering another term or signify
that the search query is in fact complete but the user has not yet
so signaled.
[0035] Regardless of the way in which the partial input is
identified, it is transmitted to the search engine 208 (308) for
processing. In response to the partial search query, the search
engine 208 returns a set of ordered predicted search queries and/or
URLs (310) which is presented to the user (312) ordered in
accordance with a ranking criteria. The predictions may be
displayed to the user in a number of ways. For example, the
predictions could be displayed in a drop-down window, a persistent,
or non-persistent window or other ways. In some embodiments,
queries which the user had previously submitted could be visually
indicated to the user (e.g., by highlighting the user's own
previously entered queries).
[0036] In some embodiments, the predicted search queries are
ordered in accordance with a frequency of submission by a community
of users. In some embodiments, the search queries are ordered, at
least in part, in accordance with a last time/date value that the
query was submitted. In some embodiments, the search queries are
ordered in accordance with personalization information, such as
user personalization information or community information. For
instance, user personalization information may include information
about subjects, concepts or categories of information that are of
interest to the user. The user personalization information may be
provided directly by the user, or may be inferred with the user's
permission from the user's prior search or browsing activities, or
may be based at least in part on information about a group
associated with the user or to which the user belongs (e.g., as a
member, or as an employee). The set of predicted search queries may
be initially ordered in accordance with a first ranking criteria,
such as predefined popularity criteria, and then reordered if any
of the predicted search queries match the user personalization
information of the user so as to place the matching predicted
search queries at or closer to the top of the ordered set of
predicted search queries.
[0037] One skilled in the art will recognize a number of ways to
present the predicted search queries and/or URLs to the user. For
example, the predicted search queries and/or URLs might be
presented in a drop down menu. Regardless of the manner in which
the predicted queries and/or URLs are presented to the user, the
user may select one of the queries and/or URLs if the user
determines that one of the predictions matches the intended entry.
In some instances, the predictions may provide the user with
additional information which had not been considered. For example,
a user may have one query in mind as part of a search strategy, but
seeing the predicted results causes the user to alter the input
strategy. Once the set is presented (312), the user's input is
again monitored. If the user selects one of the predictions
(302-final), the request is transmitted either to the search engine
208 as a search request or to a resource host as a URL request
(304), as applicable. After the request is transmitted, the user's
input activities are again monitored (302). As mentioned above, in
some embodiments, the URL request is transmitted to search engine
208 for logging purposes.
[0038] If, on the other hand, the user has not selected one of the
predictions within a specified time period, then it is likely that
the user did not find a satisfactory prediction in the predictions
that were initially returned. For example, a user's intended input
did not have a high enough ranking value to be included in the set
of ordered predictions. Accordingly, in some optional embodiments,
if the user has not selected one of the predictions within a
specified period of time (e.g., 5 or 10 seconds) (302-timeout),
then a request is sent to the search engine 208 for another set of
predictions (318). The subsequent set of predictions could include
predictions having ranking values lower than the set previously
submitted. Alternately, a second set of criteria may be used to
identify predictions in the second set, where the second set of
criteria are different than a first set of criteria used to select
and rank the first set of predictions. For instance, one of the two
sets may use selection criteria that takes into account personal
information about the requestor while the other set does not. In
some optional embodiments, other triggers may be used to request
one or more sets of subsequent predictions. For example, a
user-initiated activity (e.g., pressing the "tab" key, an arrow
key, a function key, and the like) may cause a request for a
subsequent set. In some embodiments, information associated with a
search requestor is maintained at the server to identify which
predicted results have already been conveyed to the search
requestor. In some embodiments, the client includes information in
the request for a subsequent request which indicates which results
have already been conveyed to the search requestor. In one such
embodiment the prediction server 212 uses this information to
exclude from subsequently predicted results either all of the
previously predicted results or a subset of the previously
predicted results. In another embodiment, the information about
previously predicted results is used by the prediction server 212
to produce additional or different results only if the prediction
server 212 is able to identify additional predicted results that
match the requestor's partial query. In some embodiments,
triggering a subsequent set of predictions causes predictions to be
made using a search requestor's search queries stored locally,
while in other embodiments the subsequent set of predictions
includes both predictions generated based on historical queries of
a community of users and the search requestor's historical search
queries, if any, that match the requestor's partial query.
[0039] In some embodiments, one or more sets of predicted results
are cached locally at the client. When the search requestor
modifies the current query to reflect an earlier partial input
(e.g., by backspacing to remove some characters), the set of
predicted results associated with the earlier partial input is
retrieved from the client cache and again presented again to the
user instead of the partial input being sent to the search
engine.
[0040] In some embodiments, the search engine 208 may optionally
return predicted results (320). This activity may overlap with
receiving the predictions (310) and is indicated by the dashed line
to 320 in FIG. 3. The predicted results are presented (320) and the
monitoring of the user resumes (302). The presentation to the user
can be accomplished in a number of ways. For example, the results
can be displayed in a portion of a non-persistent window, a pop-up
window, or in a portion of the current display or a portion of a
user interface. The web page used for entry of the query and for
presenting predicted results may include JavaScript or other
embedded code or instructions to facilitate the display of the
predicted results and to respond to user selection of any of the
predicted results. Other ways are envisioned. The predicted results
correspond to documents or information that would have been
returned based on the request being one or more of the predicted
queries or URLs. In some embodiments, the predicted results include
snippets of the content at one or more locations corresponding to
the predicted results. In some embodiments, the predicted results
include one or more thumbnails of one or more web pages or other
content at one or more locations corresponding to the predicted
results. In some embodiments, the results are search results based
on one or more of the predicted queries. For example, in some
embodiments, the results presented (320) may be one or more
documents relevant to one or more of the predicted queries or
predicted URLs. Accordingly, the user may have predicted results
presented that match a desired request before the user finishes
entering the request (e.g., search request or URL request). In such
situations, the processing latency as viewed by the user is
effectively reduced to less than zero because the user did not have
to complete the input to obtain the desired result.
[0041] FIG. 4 illustrates the activity occurring in the search
engine 208 when it receives an input according to some embodiments.
The search engine 208 receives the input and determines whether the
input indicates a final input or a partial input (402). If the
search engine 208 determines that the received input is a final
query (402-final query) then it determines whether search results
relevant to the query are present in the cache 232 (404). If the
relevant search results are in the cache 232 (404-yes), then those
results are returned to the client 104 (406). On the other hand, if
the search results are not in the cache (404-no), then search
results relevant to the query are obtained (408), and then returned
to the client 104 (406). In some embodiments, a URL request, when
complete, is not received by the search engine 208 because the
search assistant sends the request to the resource host. In some
embodiments, the URL request is received by the search engine 208
for tracking purposes (such as storage in a URL database) and the
request is redirected to the resource host by the search engine
208.
[0042] If the search engine 208 determines that the received input
was a partial input (402-partial), then it determines a set of
ordered matches that correspond to the partial input (410), and
transmits the set to the client 104 (412). As will be explained
below, in some embodiments, the set of ordered matches sent to the
client 104 is one of many pre-computed sets of ordered matches.
Although the following operations are described in terms of a
partial query, the same techniques are equally applicable to
partial inputs of URLs. In some embodiments, the set of ordered
matches returned is relevant only to queries. In some embodiments,
the set of ordered matches is relevant to only URLs. And, in some
embodiments, the set of ordered matches is relevant to both queries
and URLs.
[0043] To aid in understanding how, according to some embodiments,
the search engine 208 determines which set of ordered matches to
return, it is helpful to begin with a description of how the
ordered sets are created and used. FIG. 5 shows a set of data
structures associated with historical queries (i.e., queries
previously submitted) used for predicting queries corresponding to
partially entered queries. A search engine or user input prediction
system may also include a parallel set of data structures
associated with historical URLs (i.e., URLs previously submitted)
used for predicting URLs corresponding to partially entered
URLs.
[0044] Referring to FIG. 5, a historical query log 502 is filtered
by one or more filters 504 to create an authorized historical
queries list 506. An ordered set builder 508 creates one or more
fingerprint-to-table maps 510 from the authorized historical
queries list 506 based on certain criteria. When the partial query
is transmitted (FIG. 3, 308), it is received at the search engine
208 as partial query 513. A hash function 514 is applied to the
partial query 513 to create a fingerprint, i.e., a b-bit binary
value (e.g., a 64-bit number). An applicable fingerprint-to-table
map 510 (e.g., 510-1) is searched for the fingerprint (e.g., 515)
to identify a query completion table 516 associated with the
fingerprint. The query completion table 516 provides an ordered set
of predicted queries relevant to the partial query 513.
[0045] An applicable fingerprint-to-table map 510 may be selected
based on a number of different factors associated with a user or a
request. Information used to select the applicable
fingerprint-to-table map 510 could come from profile information
provided by the user or the search assistant 204, information
gleaned from the request itself (e.g., language), information
associated with the user in user information processing module 222,
or other sources. For example, fingerprint-to-table maps could be
selected based on certain connection information associated with
the user or the search requestor (e.g., device-type,
connection-speed, connection type, and the like). In some
embodiments, the number of predictions or length of each of the
query predictions depends on such connection information. Devices
with small user interfaces might receive fewer numbers of
predictions and/or queries with fewer number of terms. A query term
could have an importance factor associated with it and terms having
lower importance factors could be truncated from the query before
terms having higher importance factors. In some embodiments,
different sets of fingerprint-to-table maps 510 may be used for
respective categories of users, thereby providing predicted results
that are biased in accordance with one or more categories or topics
associated with the user. For instance, partial search queries
received from a particular website might be mapped to predicted
results using a set of fingerprint-to-table maps that were
generated from historical queries received from the same website,
or from a group of websites deemed to be similar to the particular
website. Similarly, an individual user may, with his/her
permission, have a user profile that specifies information about
the user or about a group associated with the user, and that
"personalization information" may be used to identify a respective
set of fingerprint-to-table maps for use when predicting results
for that user. It is noted that the overhead associated with adding
multiple sets of fingerprint-to-table maps 510 may be modest,
because multiple sets of fingerprint-to-table maps 510 could point
to the same query completion table 516, and the query completion
tables 516 occupy much more storage than the fingerprint-to-table
maps 516.
[0046] In some embodiments, some preprocessing occurs to the
partial query before the fingerprint is created. In one embodiment,
conspicuously misspelled words in the partial query are identified
and corrected by comparing one or more of the complete search terms
with entries in a dictionary. One or more predicted results from
queries including the correctly spelled word are merged with the
predicted results returned to the user. In another example, common
prefix information could be removed (e.g., "http://" or "www."). In
some embodiments, the terms in the query are analyzed to extract
concepts embodied in the search terms indicating a particular
category of information (e.g., "technology, "food", "music" or
"animals"). One or more predicted results from queries related to
one or more of the extracted concepts are merged with the predicted
results returned to the user.
[0047] The historical query log 502 contains a log of previously
submitted queries received by the search engine 208 over a period
of time. In some embodiments, the queries are from a particular
user. In some embodiments, the queries are from a community of
users sharing at least one similar characteristic such as belonging
to the same workgroup, using the same language, having an internet
address associated with the same country or geographic region, or
the like. The selection of the community determines the pool of
previously submitted queries from which the predictions are drawn.
Different communities would tend to produce different sets of
predictions.
[0048] The historical query log 502 may also contain information
associated with each submitted query. In some embodiments, the
query information includes the date and time that the query was
submitted or received. In some embodiments, the query information
includes the internet protocol (IP) address from where the query
was submitted. In some embodiments, the query information contains
a unique source identifier for the query (e.g., a value from a
cookie stored on the user's machine where the value is associated
with a particular search assistant 204). While the unique
identifier does not directly identify any particular user, it may
be associated with a particular installation of a browser or
toolbar. In some embodiments, a user may permit direct
identification with the unique identifier for certain
personalization features which could be accessed using user
information processing module 222.
[0049] In some embodiments, a fingerprint value is associated with
the query. The fingerprint value may be calculated by applying a
hash function to the query string. In some embodiments, other types
of meta-data are associated and stored with the query such as the
query language or other information which might be provided by the
user or search assistant in accordance with user preferences (e.g.,
identification or profile information indicating certain
preferences of the user). In some embodiments, the meta-information
includes category or concept information gleaned from analyzing the
terms in the query. The period of time over which the queries are
logged is a variable and represents a tradeoff between storage
capacity and potential accuracy of the predictions. It is likely
that longer periods of time will more accurately reflect a query's
popularity over the entire community, however, this requires more
storage. On the other hand, a popularity ranking over a long period
of time may not reflect a transient popularity for current
events.
[0050] One or more filters 504 are used to determine queries
authorized for further processing. For example, filters can
eliminate certain queries based on various criteria. In some
embodiments, a privacy filter 504 prevents queries which have not
been received from more than a certain number of unique submitters
to be included in the authorized historical queries list 506. This
could be accomplished by examining the unique identifier associated
with each query, if one exists, and identifying only those queries
which have been submitted by at least n unique submitters, where n
is a number chosen based on privacy concerns (e.g., three or five
unique submitters). In some embodiments, the filters 504 include a
filter that eliminates queries which are infrequently submitted and
therefore not likely to be selected by a user. In some embodiments,
the filters 504 include an appropriateness filter 504 that blocks
certain queries from inclusion based on a number of different
factors such as the presence of one or more particular keywords in
a query, and/or based on the content of the search results or
documents that correspond to the query. Other types of filters
could be easily imagined. For example, a filter could block queries
submitted earlier than a particular historical point in time, such
that the authorized historical queries list 506 represent recently
submitted queries. What is considered recent depends on the
embodiment (e.g., hours, days, weeks, months, or years). In yet
another example, an anti-spoofing filter 504 could be use to
prevent the query/URL prediction system from being spoofed by a
large number of a artificially generated queries or URL
submissions. For instance, an anti-spoofing filter 504 might filter
out multiple submissions of the same query or URL received from the
same user or from the same client computer.
[0051] After the historical query log 502 has been filtered by the
one or more filters 504, the result is the authorized historical
queries list 506, i.e., a list of queries eligible to be returned
to the user as suggested query completions. The authorized
historical queries list 506 includes historical query 506-1 to
historical query 506-q, where q represents the number of queries
included in the authorized historical queries list 506. The value
of q could be equal to or less than the total number of queries
filtered from the historical query log 502. For example, filtered
queries having frequencies less than a predetermined threshold
could be ignored. In some embodiments, a new authorized historical
queries list 506 is built periodically such as hourly, nightly,
weekly or other periods. In some embodiments, the current
authorized historical queries list 506 is updated based on recent
entries to the query log 224, after applicable filtering.
[0052] Each query in authorized historical queries list 506 (e.g.,
506-1) includes the query, its frequency and, optionally,
meta-information. The query could be a string of characters. The
frequency information indicates how many times the query was
submitted over a period of time. As mentioned above, a unique
identifier may be used to count the number of times unique
searchers submitted the query. Because different users may use
multiple search assistants or some queries may not include a unique
identifier, the frequency number may not represent the actual
number of unique users submitting the search query. Nonetheless, a
query's frequency can act as a proxy for a query's popularity. In
some embodiments, the authorized historical queries list 506 is
ordered alphabetically based on the query. In other embodiments,
the authorized historical queries list 506 is ordered based on the
query frequency.
[0053] The meta-information, may include information similar to the
meta-information discussed above in reference to the historical
query log 502 (e.g., location or language information). In some
instances, the same query will have entries in the historical query
log 502 which differ not in the query string, but in the
meta-information. Accordingly, the meta-information for a
particular authorized historical query 506-1 may indicate differing
meta-information for the same query. For example, the
meta-information for a query submitted from two different
locations, such as Europe or Asia, would indicate both locations as
a source location of the query. The meta-information could also
indicate user profiling information to indicate what types of users
had submitted the query. One of ordinary skill in the art will
recognize various types of meta-information that might be useful to
categorize or group queries related by common set of
characteristics (e.g., language or location). In some embodiments,
the query terms are analyzed and associated with certain categories
of information. For example, a search query including "dog" and
"breed" is associated with a "dog" or "animal" category. The
meta-information in some embodiments, contains this category
information. In some embodiments, meta-information for a single
entry in the authorized historical queries list 506 is produced
from the multiple queries, for example, by providing the date/time
of the query as the last date/time value that the query was
submitted.
[0054] The ordered set builder 508 uses the authorized historical
queries list 506 to build a set of fingerprint-to-table maps 510-1
to 510-t, where t represents the number of fingerprint-to-table
maps 510 built. Any number of fingerprint-to-table maps 510 could
be built depending on the number of ways desired to categorize
predicted queries. Each of the fingerprint-to-table maps 510
contain sets of ordered predictions each mapped to a particular
partial query. The fingerprint-to-table maps 510 differ based on
characteristics of information such as might be found in the
meta-information. For example, there may be one
fingerprint-to-table map 510 for each language (e.g., one for
English language queries; one of French language queries; one for
Japanese language queries). Similarly, different
fingerprint-to-table maps 510 could be created for geographical
regions. As another example, different fingerprint-to-table maps
510 could be created from queries from particular IP addresses or
groups of addresses, such as those from a particular network or a
particular group of individuals (e.g., a corporation). Using the
meta-information to create different fingerprint-to-table maps 510,
allows the predictions to be based on users having characteristics
similar to that of the searcher and which should increase the
likelihood of a correct prediction. In some embodiments, different
fingerprint-to-table maps 510 are based on different ranking
criteria for the queries (e.g., frequency, last date/time,
personalization categories or characteristics, and so on). In some
embodiments, different fingerprint-to-table maps 510 are based on
the type of user input (i.e., query string or URL).
[0055] Using fingerprint-to-table map 510-1 as an example, each of
the fingerprint-to-table maps 510 includes a number of entries
512-1 to 512-f, where f represents the number of entries in the
fingerprint-to-table map 510-1. The number of entries in any
particular fingerprint-to-table map 510 depends on the number of
different partial queries for which the prediction server 212 will
return predictions.
[0056] Each of the entries in the fingerprint-to-table map 510-1
(e.g., 512-2) includes a fingerprint (e.g., fingerprint (2) 515)
and a query completion table (e.g., query completion table (2)
516). The fingerprint-to-table maps 510 serve to associate
fingerprints (e.g., fingerprint (2) 515) to query completion tables
(e.g., query completion table (2) 516)).
[0057] The fingerprint (2) 515 represents a fingerprint value for a
partial query. The fingerprint (2) 515 may be calculated, for
example, by applying a hash function to a partial query to create a
b-bit binary value (e.g., a 64-bit number). Accordingly, the
fingerprint-to-table map 510-1 may be searched for a fingerprint
which matches the fingerprint of the partial query 513 (e.g.,
fingerprint 515).
[0058] The query completion table (2) 516 contains a list of query
completion fingerprints 518-1 to 518-n, where n represents the
number of query completion fingerprints in the query completion
table (2) 516. In some embodiments, n represents the number of
predicted queries returned to the search assistant 204 (e.g., 10
predicted queries). In other embodiments, less than n are returned.
In some embodiments, n is greater than the number of results to be
returned in a set of ordered queries. In some embodiments, n is
twice the number to be returned and the first n/2 are provided as a
first set of ordered predicted queries and the second n/2 are
provided as a subsequent set of ordered predicted queries (e.g.,
the second set of 10 predicted queries is sent subsequent to the
first set of 10 upon certain conditions). In some embodiments, the
query completion table 516 includes a score for each query
completion fingerprint 518. The scores are used to order the items
in the query completion table 516, in descending score order. In
some embodiments, the scores are a permanent part of the query
completion table, while in other embodiments the scores are deleted
or not kept after the formation of the query completion tables 516
is completed.
[0059] Each query completion fingerprint 518 is a fingerprint value
associated with a complete query. The query completion fingerprint
518 (e.g., 518-2) maps to an associated query record 520. The query
record 520 includes a query string 522 which contains the query
string for the complete query. This approach facilitates entries in
multiple query completion tables 516 referencing the same query
string 522, yet only requiring that the actual query string be
stored in a single location (e.g., query string 522). In some
embodiments, however, the query strings 522 may be stored in place
of the query completion fingerprints 518 in a query completion
table 516. In some embodiments, query record 520 for URL strings
include a URL title 524 representing a title associated with the
URL. In some embodiments, additional information associated with a
URL is provided in information 526.
[0060] In some embodiments, the query completion table 516 is an
ordered list of n queries relevant to the partial query associated
with the fingerprint 515. The list may be ordered in accordance
with various ranking criteria such as frequency, date/time of
submission, and so on. In some embodiments, the ranking criteria
may take into account two or more factors, such as both frequency
and date/time or submission, by generating a score or rank for each
query that takes into account each of the two or more factors. In a
simple example, historical queries whose date/time is more than 24
hours in the past may contribute a value of "1" to the ranking
score of the query, while historical queries whose date/time is
within the last 24 hours may contribute a value of "2" to the
ranking score of the query. In this example, recent historical
queries are weighted more heavily than older historical queries in
determining the rank of each authorized historical query.
[0061] In some embodiments, the ordered set builder 506 creates or
updates the fingerprint-to-table maps 510 and associated query
completion tables 516 and/or 910 (FIG. 9) periodically (e.g.,
hourly, daily, weekly) so as to keep the query and/or URL
predictions produced by the prediction server consistent with
queries and/or URLs recently submitted by the applicable community
of users.
[0062] Referring to FIG. 6, a partial query of "ho" 602 might have
a set of completed queries 604 as being relevant to the partial
query 602. The first position of the set of completed queries 604
includes the query having the highest frequency value (e.g.,
"hotmail"), it is followed in the second position with the query
having the next highest frequency value (e.g., "hot dogs"), and so
on. In this example, a complete query's relevancy to a given
partial query is determined by the presence of the partial query at
the beginning of the complete query (e.g., the characters of "ho"
begin the complete queries of "hotmail" and "hotels in San
Francisco"). In other embodiments, the relevancy is determined by
the presence of the partial query at the beginning of a search term
located anywhere in the complete query, as illustrated by the set
of completed queries 606 (e.g., the characters "ho" are found at
the beginning of "hotmail" and at the beginning of the second
search term in "cheap hotels in Cape Town").
[0063] To create the set of query completion tables 516, one of the
queries in the authorized historical queries 506 is selected (FIG.
7, 702). In some embodiments, only queries having the desired
meta-information are processed (e.g., queries in the English
language). The first partial query is identified from the selected
query (704). In one embodiment, the first partial query is the
first character of the selected query (i.e., "h" for a query string
of "hot dog ingredients"). In some embodiments, preprocessing is
applied before partial queries are identified (e.g., stripping off
"http://" or "www."). An entry is made in a table which indicates
the partial query, the complete query corresponding to the partial
query and its frequency. In other embodiments, other information
which is used for ranking is stored (e.g., date/time values, or a
ranking score computed based on two or more factors). If the
partial query does not represent the entire query, then the query
processing is not complete (708-no). Accordingly, the next partial
query is identified (710). In some embodiments, the next partial
query is identified by adding the next additional character to the
partial query previously identified (i.e., "ho" for a query string
of "hot dog ingredients"). The process of identifying (710) and of
updating of a query completion table (706) continues until the
entire query is processed (708-yes). If all of the queries have not
yet been processed (712-no), then the next query is selected and
processed until all queries are processed (712-yes). In some
embodiments, as items are added to a query completion table, the
items are inserted so that the items in the table are ordered in
accordance with the rank or score. In another embodiment, all the
query completion tables are sorted at the end of the table building
process so that the items in each query completion table are
ordered in accordance with the rank or score of the items in the
query completion table. In addition, one or more query completion
tables may be truncated so that the table contains no more than a
predefined number of entries.
[0064] Referring to FIG. 8, an exemplary processing of the first
five characters of the query string of "hot dog ingredients" is
illustrated in table 802 at 804 through 812. An exemplary
processing of the first four characters of the query string of
"hotmail" is illustrated at 814 through 820.
[0065] In some embodiments, a query completion table for a given
partial query is created by identifying the n most frequently
submitted queries relevant to the given partial query from the
table and placing them in ranked order such that the query having
the highest rank (e.g., the highest ranking score or frequency) is
at the top of the list. For example, a query completion table for
the partial query "hot" would include both complete query strings
of 808 and 818. When the ranking is based on frequency, the query
string for "hotmail" would appear above the query string for "hot
dog ingredients" because the frequency of the query string in 818
(i.e., 300,000) is larger than that of the query string in 808
(i.e., 100,000). In some embodiments, a URL's popularity could be
given a value assigned to a particular web page providing an
indication of its importance among a set of web pages (e.g.,
PageRank). Accordingly, when the ordered set of prediction is
returned to the user, the queries having a higher likelihood of
being selected are presented first. As mentioned above, other
values could be used for ranking drawn from the meta-information
(e.g., date/time values, or personalization information).
[0066] Referring to FIGS. 9 and 10, in some embodiments the number
of query completion tables is reduced by dividing the historical
query strings into "chunks" of a predefined size C, such as 4
characters. The query completion tables for partial queries of
length less than C remain unchanged. For partial queries whose
length is at least C, the partial query is divided into two
portions: a prefix portion and a suffix portion. The length of the
suffix portion, S, is equal to the length of the partial query (L)
modulo C:
[0067] S=L modulo C.
where L is the length of the partial query. The length of the
prefix portion, P, is the length of the partial query minus the
length of the suffix: P=L-S. Thus, for example, a partial query
having a length of 10 characters (e.g., "hot potato"), would have a
suffix length S of 2 and a prefix length P of 8 when the chunk size
C is 4.
[0068] When performing the process shown in FIG. 7, step 706,
identifying or creating a query completion table corresponding to a
partial query is conceptually illustrated in FIG. 9. FIG. 9
schematically illustrates the process used both for generating
query completion tables as well as for lookup when processing a
user entered partial query. When the length of the partial query is
less than the size of one "chunk", C, the partial query is mapped
to a query fingerprint 515, for example by using a hash function
514 (FIG. 5). The fingerprint 515 is mapped to a query completion
table 516 by a fingerprint to table map 510, which in turn contains
query completion fingerprints 518 or pointers to a set of query
records 520 (which contain query strings 522, FIG. 5).
[0069] When the length of the partial query is at least the size of
one chunk, C, the partial query 902 is decomposed into a prefix 904
and suffix 906, whose lengths are governed by the chunk size, as
explained above. A fingerprint 908 is generated for the prefix 904,
for example by applying a hash function 514 to the prefix 904, and
that fingerprint 908 is then mapped to a "chunked" query completion
table 910 by a fingerprint to table map 510. The structure of the
chunked query completion table 910 is different from the query
completion table 516 shown in FIG. 5, in that each entry 911 of the
chunked query completion table 910 has a suffix entry 914 as well
as a query completion fingerprint 912. Each entry 911 may
optionally include a score 916 as well, used for ordering the
entries in the query completion table 910. The suffix has a length,
S, which can be anywhere from zero to C-1, and comprises the zero
or more characters of the partial query that are not included in
the prefix 904. In some embodiments, when generating the query
completion table entries 911 for a historical query, only one entry
is made in each chunked query completion table 910 that corresponds
to the historical query. In particular, that one entry 911 contains
the longest possible suffix for the historical query, up to C-1
characters long. In other embodiments, up to C entries are made in
each chunked query completion table 910 for a particular historical
query, one for each distinct suffix.
[0070] FIG. 10 shows a set of query completion tables which contain
entries 911 corresponding to the historical query "hot potato".
This example assumes a chunk size, C, equal to four. In other
embodiments the chunk size may be 2, 3, 5, 6, 7, 8, or any other
suitable value. The chunk size, C, may be selected based on
empirical information. The first three of the query completion
tables shown in FIG. 10, 516-1 through 516-3, are for the partial
queries "h", "ho" and "hot", respectively. The next two query
completion tables, 910-1 and 910-2 correspond to the partial
queries "hot pot" and "hot potato", respectively, having partial
query lengths of 7 and 10. Referring back to step 710 of FIG. 7,
with each iteration of the loop formed in part by step 710, the
length of the partial queries initially increases by steps of 1
character, until a length of C-1 is reached, and then the length of
the partial queries increases by steps of C characters, until the
full length of the historical query is reached.
[0071] The entries 911 of each chunked query completion table are
ordered according to the ranking values (represented by scores 916)
of the query strings identified by the query completion
fingerprints 912 in the entries 911. For partial queries having
less than C characters, the number of queries in the associated
query completion table 516 is a first value (e.g., 10 or 20), which
may represent the number of queries to return as predictions. In
some embodiments, the maximum number (e.g., a number between 1000
and 10,000) of entries 911 in each chunked query completion table
910 is significantly greater than the first value. Each chunked
query completion table 910 may take the place of dozens or hundreds
of ordinary query completion tables. Therefore, each chunked query
completion table 910 is sized so as to contain a number (p) of
entries corresponding to all or almost all of the authorized
historical queries having a prefix portion that corresponds to the
chunked query completion table, while not being so long as to cause
an undue latency in generating a list of predicted queries for a
user specified partial query.
[0072] After the query completion tables 516, 910 and
fingerprint-to-table maps 510 have been generated from a set of
historical queries, these same data structures (or copies thereof)
are used for identify a predicted set of queries corresponding to a
user entered partial query. As shown in FIG. 9, the user entered
partial query is first mapped to a query fingerprint 515 or 908, by
applying a hash function 514 either to the entire partial query 902
or to a prefix portion 904 of the partial query, as determined by
the length of the partial query. The query fingerprint 515 or 904
is then mapped to a query completion table 516 or 910 by performing
a lookup of the query fingerprint in a fingerprint-to-table map
510. Finally, an ordered set of up to N predicted queries is
extracted from the identified query completion table. When the
length of the partial query is less than the chunk size, the
ordered set of predicted queries are the top N queries in the
identified query completion table. When the length of the partial
query is equal to or longer than the chunk size, the identified
query completion table is searched for the top N items that match
the suffix of the partial query. Since the entries in the query
completion table 910 are ordered in decreasing rank, the process of
searching for matching entries begins at the top and continues
until the desired number (N) of predictions to return is obtained
(e.g., 10) or until the end of the query completion table 910 is
reached. A "match" exists when the suffix 906 of the partial query
is the same as the corresponding portion of the suffix 914 in an
entry 911. For instance, referring to FIG. 10, a one letter suffix
of <p> matches entries 911-3 and 911-4 having suffixes of
<pot> and <pal>, respectively. An empty suffix (also
called a null string) having length zero matches all entries in a
query completion table, and therefore when the suffix portion of a
partial query is a null string, the top N items in the table are
returned as the predicted queries.
[0073] As noted above, the data structures and processes for
identifying an ordered set of predicted URLs that correspond to a
partial URL are the same as the data structures and processes,
described above, for identifying an ordered set of predicted
queries that correspond to a user entered partial query. Even
though URLs and query strings may have different uses, both may be
treated as a string of characters or symbols whose value may be
predicted after partial entry by a user. In some embodiments, the
set of "historical URLs" from which a set of URL completion tables
1234 (FIG. 12) and URL fingerprint-to-table maps 1236 (FIG. 12) are
built may comprise URLs entered by a particular user or a set or
community of users. In another embodiment, the set of "historical
URLs" from which a set of URL completion tables and URL fingerprint
to table maps are built may comprise the URLs of documents stored
in a document database, such as the document database of a search
engine.
[0074] FIG. 11 illustrates a user's view when using a browser and
toolbar according to some embodiments of the invention. A browser
1102 includes a toolbar 1104 including a text entry box 1106
depicting the entry of a partial query <hot>. In response to
detecting the partial query and ultimately receiving the predicted
queries from the query server, the predictions are displayed in
display area 1108 for possible selection by the user. Similarly,
while not shown, in response to detecting user entry of a partial
URL in an address bar 1110, an ordered set of predicted URLs may be
displayed in a display area (not shown) immediately below or
adjacent the address bar 1110 for possible selection by the
user.
[0075] Referring to FIG. 12, an embodiment of a search engine 1202
that implements the methods and data structures described above
includes one or more processing units (CPU's) 1204, one or more
network or other communications interfaces 1206, a memory 1208, and
one or more communication buses 1210 for interconnecting these
components. The search engine 1202 may optionally include a user
interface 1212 comprising a display device 1214 and a keyboard
1216. Memory 1208 may include high speed random access memory and
may also include non-volatile memory, such as one or more magnetic
or optical storage disks. Memory 1208 may include mass storage that
is remotely located from CPU's 1204. Memory 1208, or memory devices
such non-volatile memory devices in memory 1208, may comprise a
computer readable storage medium. Memory 1208 or the computer
readable storage medium of memory 1208 may store the following
elements, or a subset or superset of such elements: [0076] an
operating system 1218 that includes procedures for handling various
basic system services and for performing hardware dependent tasks;
[0077] a network communication module (or instructions) 1220 that
is used for connecting the search engine 1202 to other computers
via the one or more communications interfaces 1206 (wired or
wireless), such as the Internet, other wide area networks, local
area networks, metropolitan area networks, and so on; [0078] a
query server 210 for receiving full or partial queries and
returning search results and predicted queries and predicted search
results; and [0079] a prediction server 212 for receiving a partial
query and returning a set of ordered predictions of queries or
URLs.
[0080] In some embodiments, the query server 210 includes the
following elements, or a subset of such elements: a client
communications module 216 for receiving and transmitting
information; a query receipt, processing and response module 218
for receiving and responding to full search queries; a partial
query receipt, processing and response module 220 for receiving and
responding to full search queries; a user information and
processing module 222 for accessing user information from a user
information database 1226, which includes respective user profiles
1228 for a plurality of users; a query log 224 for storing
information about previously submitted queries, and a URL log or
database 225. In some embodiments, the query server 210 includes a
subset of these modules. In some embodiments, the query server 210
includes additional modules.
[0081] In some embodiments, the prediction server 212 includes the
following elements, or a subset or superset of such elements:
[0082] a query receiving module (or instructions) 1230 for
receiving a partial query; [0083] a query/URL completion table
builder (or instructions) 1232 for generating query completion
tables 516, 910 and query fingerprint-to-table maps 510; a hash
function 514; in some embodiments, the query/URL completion table
builder 1232 may also generate URL completion tables 1234 and URL
fingerprint-to-table maps 1236; and [0084] a prediction module (or
instructions) 1238 for obtaining a set of predicted queries or
URLs.
[0085] In some embodiments, the prediction server 212 may also
include one or more of the following: [0086] a personalization
module (or instructions) 1240 for selecting the set of predicted
queries based, at least in part, on certain user profile
information; [0087] a concept module (or instructions) 1242 for
determining the concepts associated with a particular query; [0088]
a community characteristics module (or instructions) 1244 for
determining a set of characteristics associated with a community of
users; and [0089] a spelling module (or instructions) 1246 for
identifying alternative spellings of a received query or query
term.
[0090] In some embodiments, one or more of the user information
processing module 222, personalization module 1240, concept module
1242, community characteristics module 1244 and spell module 1246
are not implemented. When implemented, the user profiles 1228 of
the user information processing module 222 may contain information
suitable for selecting or ordering predicted queries or URLs. For
instance, a user profile 1228 may identify categories of
information that are of interest to a particular user. A user
profile 1228 may also contain information associated with a
community of users to which a user belongs or with which the user
is associated. The user information processing module 222 may merge
personal information with the community information to generate a
user profile 1228.
[0091] When implemented, the concept module 1242 may map historical
queries to concepts or categories of information, suitable for
matching with the information in a user profile 1228. Similarly,
the concept module 1242 may be configured to map historical URLs to
concepts or categories of information, for instance by determining
a set of primary concepts, subjects or categories of information in
the content of the documents corresponding to the historical URLs.
The concept, subject or category information identified by the
concept module 1242 may be stored in the entries of the query
completion tables or URL completion tables, or in the query records
or URL records identified by the query/URL completion tables. When
processing a partial query or URL, the set of predicted queries or
URLs may be reordered so that the predicted queries or URLs whose
concept, subject or category information matches the information in
the user profile of the requesting user are placed higher in the
list of predicted queries or URLs than those predicted queries or
URLs whose concept or category information does not match the
information in the user profile of the requesting user.
[0092] In another embodiment, the concept module 1242 may be
configured to map one or more terms in a partial query to one or
more substitute terms in accordance with a conceptual or category
mapping of those terms. An ordered set of predicted queries are
generated for a partial query containing the one or more substitute
terms, and those predicted queries are then transmitted to the
user, either separately or merged with the results produced using
the partial query as entered by the user.
[0093] FIG. 12 depicts the internal structure of a search engine
1202 in one embodiment. It should be understood that in some other
embodiments the search engine 1202 may be implemented using
multiple servers so as to improve its throughput and reliability.
For instance the query log 224 could be implemented on a distinct
server that communications with and works in conjunction with other
ones of the servers in the search engine 1202. As another example,
the query/URL completion table builder 1232 could be implemented in
separate servers or computing devices (e.g., ordered set builder
242, FIG. 2).
[0094] Although the discussion herein has been made with reference
to a search engine designed for use with documents remotely located
from the search requestor, it should be understood that the
concepts disclosed herein are equally applicable to other search
environments. For example, the same techniques described herein
could apply to queries against any type of information repository
against which queries, or searches, are run (e.g., an address book,
a product information database, a file server, a web site and so
on). Accordingly, the term "search engine" should be broadly
construed to encompass all such uses.
[0095] Referring to FIG. 13, an embodiment of a client system 1300
that implements the methods described above includes one or more
processing units (CPU's) 1302, one or more network or other
communications interfaces 1304, memory 1306, and one or more
communication buses 1308 for interconnecting these components. The
search engine 1300 may optionally include a user interface 1310
comprising a display device 1312 and/or a keyboard 1314. Memory
1306 may include high speed random access memory and may also
include non-volatile memory, such as one or more magnetic or
optical storage disks. Memory 1306 may include mass storage that is
remotely located from CPU's 1302. Memory 1306, or memory devices
such non-volatile memory devices in memory 1306, may comprise a
computer readable storage medium. Memory 1306 or the computer
readable storage medium of memory 1306 may store: [0096] an
operating system 1316 that includes procedures for handling various
basic system services and for performing hardware dependent tasks;
[0097] a network communication module (or instructions) 1318 that
is used for connecting the client system 1300 to other computers
via the one or more communications network interfaces 1304 and one
or more communications networks, such as the Internet, other wide
area networks, local area networks, metropolitan area networks, and
so on; and [0098] a browser or tool 1320 for interfacing with a
user to input search queries, and for displaying search results;
and [0099] a search assistant 1322.
[0100] In some embodiments, the search assistant 1322 is separate
from the browser/tool 1320, while in other embodiments the search
assistant is incorporated in the browser/tool 1320.
[0101] The search assistant 1322 may include the following
elements, or a subset of such elements: an entry and selection
monitoring module (or instructions) 1324 for monitoring the entry
of search queries and selecting partial queries for transmission to
the search engine; a transmission module (or instructions) 1326 for
transmitting partial search queries and final search queries to the
search engine; a predicted query receipt module (or instructions)
1328 for receiving predicted queries; a predicted search results
receipt module (or instructions) 1330 for receiving predicted
search results; display module (or instructions) 1332 for
displaying predictions and results; and optionally, a search
results receipt module (or instructions) 1334 for receiving search
results. The transmission of final (i.e., completed) queries,
receiving search results for completed queries, and displaying such
results may be handled by the browser/tool 1320, the search
assistant 1322, or a combination thereof. The search assistant 1322
may also provide a corresponding set of functions for handling
partial and complete URLs, which may be handled by either the same
elements or a parallel set of elements as those described above.
The search assistant 1322 could be implemented in many ways. For
example, the search assistant 1322 could be implemented as part of
a browser, as part of a toolbar, as part of a desktop application
or on a web page using executable instructions (such as
JavaScript). At a minimum, the search assistant transmits partial
query information to a search system. The search assistant may also
enable the display of predicted results and user selection of a
displayed predicted result.
[0102] Although illustrated in FIGS. 12 and 13 as distinct modules
or components, the various modules or components may be located or
co-located within either the search engine or the client. For
example, in some embodiments, portions of prediction server 212,
and/or the various query completion tables 516 and/or 910 are
resident on the client system 202 or form part of the search
assistant 204. For example, in some embodiments query completion
tables and fingerprint-to-table maps for the most popular searches
may be periodically downloaded to a client system 202, thereby
providing fully client-based query or URL input prediction for at
least some partially input queries or URLs.
[0103] In another embodiment, the search assistant 204 may include
a local version of the prediction server 212, for making search or
URL predictions based at least in part on prior searches and URL
entries of the user. Alternately, or in addition, the local
prediction server 212 may generate predictions based on data
downloaded from a search engine or remote prediction server.
Further, the client assistant 204 may merge locally generated and
remotely generated prediction sets for presentation to the user.
The results could be merged in any of a number of ways, for
example, by interleaving the two sets or by merging the sets while
biasing queries previously submitted by the user such that those
queries would tend to be placed or inserted toward the top of the
combined list of predicted queries. In some embodiments, the client
assistant 204 inserts queries deemed important to the user into the
set of predictions. For example, a query frequently submitted by
the user, but not included in the set obtained from the search
engine could be inserted into the predictions.
[0104] Although some of various drawings illustrate a number of
logical stages in a particular order, stages which are not order
dependent may be reordered and other stages may be combined or
broken out. While some reordering or other groupings are
specifically mentioned, others will be obvious to those of ordinary
skill in the art and so do not present an exhaustive list of
alternatives. Moreover, it should be recognized that the stages
could be implemented in hardware, firmware, software or any
combination thereof.
[0105] The foregoing description, for purpose of explanation, has
been described with reference to specific embodiments. However, the
illustrative discussions above are not intended to be exhaustive or
to limit the invention to the precise forms disclosed. Many
modifications and variations are possible in view of the above
teachings. The embodiments were chosen and described in order to
best explain the principles of the invention and its practical
applications, to thereby enable others skilled in the art to best
utilize the invention and various embodiments with various
modifications as are suited to the particular use contemplated.
* * * * *