U.S. patent application number 13/569460 was filed with the patent office on 2012-11-29 for semantic advertising selection from lateral concepts and topics.
This patent application is currently assigned to MICROSOFT CORPORATION. Invention is credited to Tarek Najm, Rajeev Prasad, Munirathnam Srikanth, Abhinai Srivastava, Arungunram Chandrasekaran Surendran, Phani Vaddadi, Viswanath Vadlamani.
Application Number | 20120303444 13/569460 |
Document ID | / |
Family ID | 44354432 |
Filed Date | 2012-11-29 |
United States Patent
Application |
20120303444 |
Kind Code |
A1 |
Vadlamani; Viswanath ; et
al. |
November 29, 2012 |
SEMANTIC ADVERTISING SELECTION FROM LATERAL CONCEPTS AND TOPICS
Abstract
Advertisements are selected for presentation on search result
pages and web pages based on phrases generated from lateral
concepts and topics identified for the search result pages and web
pages. A search query or an indication of a web page is received
for which advertisements are to be provided. Lateral concepts and
topics are identified based on the search query or content of the
web page. The lateral concepts and topics are used as phrases for
selecting advertisements from an advertisement inventory. Selected
advertisements are provided for presentation on a search results
page in response to a search query or on a web page initially
identified.
Inventors: |
Vadlamani; Viswanath;
(Redmond, WA) ; Srivastava; Abhinai; (Seattle,
WA) ; Najm; Tarek; (Kirkland, WA) ; Srikanth;
Munirathnam; (Redmond, WA) ; Vaddadi; Phani;
(Issaquah, WA) ; Surendran; Arungunram
Chandrasekaran; (Sammamish, WA) ; Prasad; Rajeev;
(Bothell, WA) |
Assignee: |
MICROSOFT CORPORATION
Redmond
WA
|
Family ID: |
44354432 |
Appl. No.: |
13/569460 |
Filed: |
August 8, 2012 |
Related U.S. Patent Documents
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
|
|
12701330 |
Feb 5, 2010 |
8260664 |
|
|
13569460 |
|
|
|
|
Current U.S.
Class: |
705/14.42 ;
705/14.54 |
Current CPC
Class: |
G06Q 30/0251 20130101;
G06F 16/951 20190101; G06Q 30/02 20130101 |
Class at
Publication: |
705/14.42 ;
705/14.54 |
International
Class: |
G06Q 30/02 20120101
G06Q030/02 |
Claims
1. One or more computer-readable media storing computer-useable
instructions that, when used by one or more computing devices,
cause the one or more computing devices to perform a method
comprising: receiving a search query or an indication of a web
page; identifying one or more lateral concepts based on content
identified as being relevant to the search query or content of the
web page, wherein each lateral concept is identified as a candidate
phrase for advertisement selection purposes; identifying one or
more topics based on the search query or the content of the web
page, wherein each topic is identified as a candidate phrase for
advertisement selection purposes; selecting one or more phrases
from the identified candidate phrases; querying an advertisement
inventory using the one or more selected phrases to select one or
more advertisements; and providing the one or more advertisements
for presentation to a user.
2. The one or more computer-readable media of claim 1, wherein
identifying one or more lateral concepts comprises: obtaining a
first set of content from storage that corresponds to the user
query or content of the web page; identifying a plurality of
categories associated with the obtained first set of content; and
selecting a subset of the plurality of identified categories as
lateral concepts.
3. The one or more computer-readable media of claim 1, wherein
identifying one or more lateral concepts comprises: calculating
similarity between content in storage and the user query or the
content of the web page; creating a collection of content having a
predetermined number of content similar to the user query or the
content of the web page; identifying a plurality of categories that
correspond to content in the collection of content; and selecting
several identified categories as lateral concepts.
4. The one or more computer-readable media of claim 1, wherein a
search query is received and wherein identifying one or more topics
comprises: determining if an ontology mapping exists for the search
query; if an ontology mapping exists for the search query,
retrieving a first set of topics based on the ontology mapping and
adding the first set of topics to a list of topics; performing a
search using the search query to obtain a plurality of search
results, each search result corresponding with a document snippet;
receiving at least a portion of the document snippets as a document
set for further analysis; comparing each document snippet in the
document set to an ontology of topics; for each document snippet in
which positive topic identification is determined, assigning the
document snippet to a corresponding topic and removing (320) the
document snippet from the document set; adding at least one topic
identified from the ontology of topics to the list of topics;
comparing each document snippet remaining in the document set to an
ontology of partial topics; for each document snippet in which
positive partial topic identification is determined, assigning the
document snippet to a corresponding partial topic and removing the
document snippet from the document set; naming at least one partial
topic having one or more assigned document snippets; adding at
least one named partial topic to the list of topics; computing
independent key-phrases from document snippets remaining in the
document set; assigning documents to independent key-phrases;
identifying at least one key-phrase topic; and adding the at least
one key-phrase topic to the list of topics.
5. The one or more computer-readable media of claim 4, wherein
naming a partial topic comprises: identifying occurrences of a
partial topic identifier word for the partial topic within one or
more document snippets assigned to the partial topic; extracting
words and/or phrases occurring around identified occurrences of the
partial topic identifier word within the one or more document
snippets; counting frequency of each extracted word and/or phrase;
selecting a most frequently used word or phrase; and naming the
partial topic using the partial topic identifier and the most
frequently used word or phrase.
6. The one or more computer-readable media of claim 5, wherein
counting frequency of each extracted word and/or phrase comprises
tracking position of each extracted word and/or phrase relative to
the partial topic identifier word, and wherein naming the partial
topic comprises sequencing the partial topic identifier word and
the most frequently used word or phrase based on position
information for the most frequently used word or phrase.
7. The one or more computer-readable media of claim 4, wherein
computing independent key-phrases from document snippets remaining
in the document set comprises: generating candidate key-phrases
from the document snippets remaining the document set; evaluating
candidate key-phrases for independence; merging mutually dependent
candidate key-phrases; and identifying a most frequent candidate
key-phrase for each group of merged mutually dependent
key-phrases.
8. The one or more computer-readable media of claim 1, wherein
selecting one or more phrases from the candidate phrases comprises:
ranking each candidate phrase based on an estimate of an extent to
which each candidate phrase will produce advertising revenue; and
selecting the one or more phrases based on ranking;
9. The one or more computer-readable media of claim 1, wherein
querying the advertisement inventory using the one or more selected
phrases to select one or more advertisements comprises performing
an auction process to select the one or more advertisements based
on relevance of each advertisement to the one or more phrases and
based on monetization factors associated with each
advertisement.
10. The one or more computer-readable media of claim 1, wherein a
search query is received, and wherein providing the one or more
advertisements for presentation to the user comprises providing the
one or more advertisements for presentation on a search results
page including search results in response to the search query.
11. The one or more computer-readable media of claim 10, wherein
the search results page includes the one or more lateral concepts
allowing the user to access content associated with the one or more
lateral concepts.
12. The one or more computer-readable media of claim 11, wherein
the search results page includes the one or more topics in a table
of contents allowing the user to select a topic from the one or
more topics to view content associated with the selected topic.
13. The one or more computer-readable media of claim 1, wherein an
indication of a web page is received, and wherein providing the one
or more advertisements for presentation to the user comprises
providing the one or more advertisements for presentation on the
web page.
14. A computer system including one or more processors and one or
more computer-readable media configured to select and deliver
advertisements, the computer system including: a phrase generator
to generate candidate phrases based on a search query or identified
web page, wherein the phrase generator includes a lateral concept
generator and a semantic topic engine, wherein the lateral concept
generator is configured to select lateral concepts from categories
associated with content in storage based on similarity scores for
the stored content, wherein the semantic topic engine is configured
to identify topics by analyzing the search query or web page with
an ontology of topics and with an ontology of partial topics and by
generating key-phrase topics, and wherein the lateral concepts and
topics are identified as candidate phrases; a phrase selection
component configured to select one or more phrases from the
candidate phrases; and an advertising delivery system including an
advertisement selection component and an advertisement delivery
engine, wherein the advertisement selection component is configured
to query an advertisement inventory using the one or more phrases
to select one or more advertisements, and wherein the advertisement
delivery engine is configured to deliver the one or more
advertisements for presentation to a user.
15. The computer system of claim 14, wherein the advertising
delivery system delivers an advertisement for presentation on a
search results page that includes search results in response to the
search query, and wherein the search results page includes lateral
concepts and a table of contents listing the topics.
16. A computer-implemented executed by a search engine, the
computer-implemented method comprising: receiving a search query or
an indication of a web page; identifying one or more lateral
concepts based on content identified as being relevant to the
search query or content of the web page, wherein each lateral
concept is identified as a candidate phrase for advertisement
selection purposes; identifying one or more topics based on the
search query or the content of the web page, wherein each topic is
identified as a candidate phrase for advertisement selection
purposes; selecting one or more phrases from the identified
candidate phrases; querying an advertisement inventory using the
one or more selected phrases to select one or more advertisements;
and providing the one or more advertisements for presentation to a
user.
17. The method of claim 16, wherein selecting one or more phrases
from the candidate phrases comprises: ranking each candidate phrase
based on an estimate of an extent to which each candidate phrase
will produce advertising revenue; and selecting the one or more
phrases based on ranking;
18. The method of claim 16, wherein querying the advertisement
inventory using the one or more selected phrases to select one or
more advertisements comprises performing an auction process to
select the one or more advertisements based on relevance of each
advertisement to the one or more phrases and based on monetization
factors associated with each advertisement.
19. The method of claim 16, wherein a search query is received, and
wherein providing the one or more advertisements for presentation
to the user comprises providing the one or more advertisements for
presentation on a search results page including search results in
response to the search query.
20. The method of claim 19, wherein the search results page
includes the one or more lateral concepts allowing the user to
access content associated with the one or more lateral concepts and
the search results page includes the one or more topics in a table
of contents allowing the user to select a topic from the one or
more topics to view content associated with the selected topic.
Description
CROSS-REFERENCE TO RELATED APPLICATIONS
[0001] This application is a continuation of U.S. application Ser.
No. 12/701,330, entitled "Semantic Advertising Selection From
Lateral Concepts and Topics," filed Feb. 5, 2010, which is
incorporated in this application by reference. Also, this
application is related by subject matter to the invention disclosed
in the following U.S. patent applications: U.S. Pat. No. 8,150,859
(Attorney Docket Number MFCP.153201), entitled "Semantic Table of
Contents for Search Results;" and U.S. application Ser. No.
12/700,980 (Attorney Docket Number MFCP.153202), entitled
"Generating and Presenting Lateral Concepts;" each of which is
assigned or under obligation of assignment to the same entity as
this application, and incorporated in this application by
reference.
BACKGROUND
[0002] Online advertising has become a significant aspect of the
web browsing experience. A number of advertising delivery systems
currently operate to select and deliver contextual advertisements
for placement on, for instance, web pages and search result pages.
In the context of web pages, advertising delivery systems operate
to analyze the text of the web pages to identify keywords that are
used for selecting advertisements for placement on the web pages.
In the context of search, when a user submits a search query to a
search engine, keywords are identified based on the terms of the
search query and/or based on content of the search results. The
keywords are used for selecting advertisements that are presented
in conjunction with general search results for the user's
query.
[0003] Typically, advertising delivery system providers receive
payment from advertisers based upon pay-per-performance models
(e.g., cost-per-click or cost-per-action models). In such models,
the advertisements returned with search results for a given search
query include links to landing pages that contain the advertisers'
content. A search engine provider receives payment from an
advertiser when a user clicks on the advertiser's advertisement to
access the landing page and/or otherwise performs some action after
accessing the landing page (e.g., purchases the advertiser's
product).
[0004] In the pay-per-performance model, advertising delivery
systems select advertisements for web pages and search queries
based on monetization. In other words, advertisements are selected
to maximize advertising revenue. This is often performed through an
auction process. Advertisers bid for particular words and/or
phrases as a way for selecting advertisements and determining the
order in which advertisements will be displayed for a given web
page or search query. Bids are typically made as cost-per-click
commitments. That is, the advertiser bids a dollar amount it is
willing to pay each time a user selects or clicks on a displayed
advertisement selected and presented as part of a web page or a
result of a given search query.
[0005] In some instances, analysis of some web pages and search
queries may only identify keywords that have not been bid on by
advertisers or may result in only minimal keywords that have been
bid on by advertisers. As a result, only minimal or no
advertisements are selected for presentation on these web pages or
in conjunction with search results for these search queries.
SUMMARY
[0006] This summary is provided to introduce a selection of
concepts in a simplified form that are further described below in
the Detailed Description. This summary is not intended to identify
key features or essential features of the claimed subject matter,
nor is it intended to be used as an aid in determining the scope of
the claimed subject matter.
[0007] Embodiments of the present invention relate to using lateral
concepts and topics identified for search queries and web pages as
phrases for selecting advertisements for presentation on search
result pages and web pages. In the context of search, when a search
query is received, lateral concepts and topics are identified for
the search query. The lateral concepts and topics are used as
phrases for advertisement selection. Selected advertisements are
provided on a search results page with search results in response
to the search query. In the context of a web page, when an
indication of a web page is received, lateral concepts and topics
are identified for the web page. The lateral concepts and topics
are used as phrases for advertisement selection. Selected
advertisements are provided on the web page.
BRIEF DESCRIPTION OF THE DRAWINGS
[0008] The present invention is described in detail below with
reference to the attached drawing figures, wherein:
[0009] FIG. 1 is a block diagram of an exemplary computing
environment suitable for use in implementing embodiments of the
present invention;
[0010] FIG. 2 is a block diagram of an exemplary system for
delivering advertisements in accordance with an embodiment of the
present invention;
[0011] FIG. 3 is a block diagram of an exemplary lateral concept
generator in accordance with an embodiment of the present
invention;
[0012] FIG. 4 is a block diagram of an exemplary semantic topic
engine in accordance with an embodiment of the present
invention;
[0013] FIG. 5 is a flow diagram showing an overall method for
selecting advertisements in accordance with an embodiment of the
present invention;
[0014] FIG. 6 is a flow diagram showing a method for generating
lateral concepts for use in selecting advertisements in accordance
with an embodiment of the present invention;
[0015] FIG. 7 is a flow diagram showing another method for
generating lateral concepts for use in selecting advertisements in
accordance with an embodiment of the present invention;
[0016] FIG. 8 is a flow diagram showing a method for identifying
topics for use in selecting advertisements in accordance with an
embodiment of the present invention;
[0017] FIG. 9 is a flow diagram showing a method for naming a
partial topic in accordance with an embodiment of the present
invention;
[0018] FIG. 10 is a flow diagram showing a method for computing
independent key-phrases in accordance with an embodiment of the
present invention;
[0019] FIGS. 11A-11C include a flow diagram showing a method for
determining topics for a search query for use in selecting
advertisement in accordance with an embodiment of the present
invention;
[0020] FIG. 12 is an illustrative screen display showing a search
results page having advertisements selected in accordance with an
embodiment of the present invention; and
[0021] FIG. 13 is an illustrative screen display showing a web page
having advertisements selected in accordance with an embodiment of
the present invention.
DETAILED DESCRIPTION
[0022] The subject matter of the present invention is described
with specificity herein to meet statutory requirements. However,
the description itself is not intended to limit the scope of this
patent. Rather, the inventors have contemplated that the claimed
subject matter might also be embodied in other ways, to include
different steps or combinations of steps similar to the ones
described in this document, in conjunction with other present or
future technologies. Moreover, although the terms "step" and/or
"block" may be used herein to connote different elements of methods
employed, the terms should not be interpreted as implying any
particular order among or between various steps herein disclosed
unless and except when the order of individual steps is explicitly
described.
[0023] Embodiments of the present invention are generally directed
to identifying phrases for advertisement selection based on lateral
concepts and topics identified for search queries and web pages. In
the context of search, when a search query is received, lateral
concepts and topics are identified for the search query and used as
phrases for selecting advertisements for placement on a search
results page in response to the search query. In the context of a
web page, when an identification of a web page is received, lateral
concepts and topics are identified for the web page and used as
phrases for selecting advertisements for placement on the web
page.
[0024] As used herein, the term "lateral concept" refers to words
or phrases that represent orthogonal topics of a search query or
web page. The lateral concepts provide concepts that are orthogonal
to a received search query (and/or content corresponding to the
search query) or to an identified web page on which advertisements
are to be placed. In some embodiments, the lateral concepts may be
stored in an index with a pointer to one or more search queries
and/or web pages. Accordingly, the stored lateral concepts may be
identified in response to subsequent search queries--similar to
previous queries--received at a search engine or in response to web
page requests.
[0025] In some embodiments in which lateral concepts are determined
for a search query, a search results page provided in response to a
search query also includes an indication of identified lateral
concepts. The lateral concepts allow a user to navigate a large
collection of content having structured data, semi-structured data,
and unstructured data. A computer system generates the lateral
concepts by processing the collection of content matching the
search query provided by the user and selecting categories for the
content. The lateral concepts comprise a subset of the selected
categories. The lateral concepts are presented to the user along
with search results matching the search query. The lateral concepts
allow the search engine to provide concepts that are orthogonal to
a search query or content corresponding to the search query. In
turn, the user may select one of the lateral concepts to search the
combination of structured, unstructured, and semi-structured data
for content corresponding to the selected lateral concept.
[0026] For instance, a search engine may receive a search query for
Seattle Space Needle from a user. The search engine processes the
query to identify lateral concepts and search results. The lateral
concepts may be selected from the structure of metadata stored with
content for Seattle Space Needle and/or the lateral concepts may be
selected from feature vectors generated by parsing search results
associated with the search query.
[0027] The storage structure may include metadata, e.g., content
attributes for the Seattle Space Needle. The Seattle Space Needle
content attributes may include a tower attribute, a Seattle
attraction attribute, and an architecture attribute. The tower
attribute may include data that specifies the name and height of
the Seattle Space Needle and other towers, such as Taipei 101,
Empire State Building, Burj, and Shanghai World Financial Center.
The Seattle attraction attribute may include data for the name and
location of other attractions in Seattle, such as Seattle Space
Needle, Pike Place Market, Seattle Art Museum, and Capitol Hill.
The architecture attribute may include data for the architecture
type, modern, ancient, etc., for each tower included in the tower
attribute. Any of the Seattle Space Needle content attributes may
be returned as a lateral concept by the search engine.
[0028] Alternatively, the search results may be processed by a
computer system to generate lateral concepts that are returned with
the search results. The content associated with the search results
is parsed to identify feature vectors. The feature vectors include
a category element that is associated with the content. The feature
vectors are used to compare the search results and calculate a
similarity score between the search results or between the search
results and the query. The categories in the feature vectors are
selected by the computer system based on the similarity score and
returned as lateral concepts in response to the search query.
[0029] As noted above, topics may also be identified for search
queries and web pages and used as phrases for advertisement
selection. Topics may be identified in a number of different
manners within various embodiments of the present invention. In
some embodiments, when a search query or identification of a web
page is received, it is determined whether an ontology mapping
already exists for the search query or web page. For instance, a
number of topics may have been manually or algorithmically
generated and cached for a search query matching the received
search query or for the identified web page. In such embodiments,
topics from the existing ontology mapping are retrieved. In further
embodiments, the search query or identified web page is analyzed in
conjunction with an ontology of topics and/or an ontology of
partial topics to identify relevant topics. In still further
embodiments, the search query or identified web page is analyzed to
identify independent key-phrases, and key-phrase topics are
selected. When a large number of topics are identified, the topics
are ranked, and the highest ranking topics are selected.
[0030] Selected topics may be used as phrases for advertisement
selection. In some embodiments, selected topics may also be used to
generate a table of contents for search results in response to a
search query. When a search query is received, search results are
retrieved. Additionally, topics relevant to the search query and
search results are identified, and a table of contents is generated
from identified topics. A search results page is returned in
response to the search query that includes search results and the
generated table of contents. A user may select topics from the
table of contents to view different search results relevant to each
topic. In some embodiments, the table of contents is static as the
user selects different topics from the table of contents to view
different sets of search results, thereby allowing the user to
navigate search results within the context of the initial search
query.
[0031] Accordingly, in one aspect, an embodiment of the present
invention is directed to one or more computer-readable media
storing computer-useable instructions that, when used by one or
more computing devices, cause the one or more computing devices to
perform a method. The method includes receiving a search query or
an indication of a web page. The method also includes identifying
one or more lateral concepts based on the search query or content
of the web page, wherein each lateral concept is identified as a
candidate phrase for advertisement selection purposes. The method
further includes identifying one or more topics based on the search
query or the content of the web page, wherein each topic is
identified as a candidate phrase for advertisement selection
purposes. The method also includes selecting one or more phrases
from the identified candidate phrases. The method further includes
querying an advertisement inventory using the one or more selected
phrases to select one or more advertisements. The method still
further includes providing the one or more advertisements for
presentation to a user.
[0032] In another embodiment, an aspect of the invention is
directed to a computer system including one or more processors and
one or more computer-readable media configured to select and
deliver advertisements. The computer system includes a phrase
generator to generate candidate phrases based on a search query or
identified web page. The phrase generator includes a lateral
concept generator and a semantic topic engine. The lateral concept
generator is configured to select lateral concepts from categories
associated with content in storage based on similarity scores for
the stored content. The semantic topic engine is configured to
identify topics by analyzing the search query or web page with an
ontology of topics and with an ontology of partial topics and by
generating key-phrase topics. The lateral concepts and topics are
identified as candidate phrases. The computer system also includes
a phrase selection component configured to select one or more
phrases from the candidate phrases. The computer system further
includes an advertising delivery system including an advertisement
selection component and an advertisement delivery engine. The
advertisement selection component is configured to query an
advertisement inventory using the one or more phrases to select one
or more advertisements. The advertisement delivery engine is
configured to deliver the one or more advertisements for
presentation to a user.
[0033] A further embodiment of the present invention is directed to
one or more computer-readable media storing computer-useable
instructions that, when used by one or more computing devices,
cause the one or more computing devices to perform a method. The
method includes receiving a search query and performing a search
using the search query to identify content from storage that
corresponds with the search query. The method also includes
identifying a plurality of categories associated with the obtained
content and selecting a subset of the plurality of identified
categories as lateral concepts, wherein the lateral concepts are
identified as candidate phrases for advertisement selection. The
method further includes receiving a plurality of documents snippets
from the search. The method also includes identifying a first set
of one or more candidate topics by comparing one or more document
snippets to an ontology of topics, identifying a second set of one
or more candidate topics by comparing one or more document snippets
to an ontology of partial topics, and identifying a third set of
one or more candidate topics by generating key-phrase topics from
one or more document snippets. The method further includes
selecting topics from the first, second, and third set of candidate
topics as candidate phrases for advertisement selection. The method
still further includes selecting one or more phrases from the
identified candidate phrases, querying an advertisement inventory
using the one or more selected phrases to select one or more
advertisements, and providing the one or more advertisements for
presentation to a user.
[0034] Having briefly described an overview of embodiments of the
present invention, an exemplary operating environment in which
embodiments of the present invention may be implemented is
described below in order to provide a general context for various
aspects of the present invention. Referring initially to FIG. 1 in
particular, an exemplary operating environment for implementing
embodiments of the present invention is shown and designated
generally as computing device 100. Computing device 100 is but one
example of a suitable computing environment and is not intended to
suggest any limitation as to the scope of use or functionality of
the invention. Neither should the computing device 100 be
interpreted as having any dependency or requirement relating to any
one or combination of components illustrated.
[0035] The invention may be described in the general context of
computer code or machine-useable instructions, including
computer-executable instructions such as program modules, being
executed by a computer or other machine, such as a personal data
assistant or other handheld device. Generally, program modules
including routines, programs, objects, components, data structures,
etc., refer to code that perform particular tasks or implement
particular abstract data types. The invention may be practiced in a
variety of system configurations, including hand-held devices,
consumer electronics, general-purpose computers, more specialty
computing devices, etc. The invention may also be practiced in
distributed computing environments where tasks are performed by
remote-processing devices that are linked through a communications
network.
[0036] With reference to FIG. 1, computing device 100 includes a
bus 110 that directly or indirectly couples the following devices:
memory 112, one or more processors 114, one or more presentation
components 116, input/output ports 118, input/output components
120, and an illustrative power supply 122. Bus 110 represents what
may be one or more busses (such as an address bus, data bus, or
combination thereof). Although the various blocks of FIG. 1 are
shown with lines for the sake of clarity, in reality, these blocks
represent logical, not necessarily actual, components. For example,
one may consider a presentation component such as a display device
to be an I/O component. Also, processors have memory. We recognize
that such is the nature of the art, and reiterate that the diagram
of FIG. 1 is merely illustrative of an exemplary computing device
that can be used in connection with one or more embodiments of the
present invention. Distinction is not made between such categories
as "workstation," "server," "laptop," "hand-held device," etc., as
all are contemplated within the scope of FIG. 1 and reference to
"computing device."
[0037] Computing device 100 typically includes a variety of
computer-readable media. Computer-readable media can be any
available media that can be accessed by computing device 100 and
includes both volatile and nonvolatile media, removable and
non-removable media implemented in any method or technology for
storage of information such as computer-readable instructions, data
structures, program modules or other data. Computer-readable media
includes, but is not limited to, RAM, ROM, EEPROM, flash memory or
other memory technology, CD-ROM, digital versatile disks (DVD) or
other optical disk storage, magnetic cassettes, magnetic tape,
magnetic disk storage or other magnetic storage devices, or any
other medium which can be used to store the desired information and
which can be accessed by computing device 100. Combinations of any
of the above should also be included within the scope of
computer-readable media.
[0038] Memory 112 includes computer-storage media in the form of
volatile and/or nonvolatile memory. The memory may be removable,
nonremovable, or a combination thereof. Exemplary hardware devices
include solid-state memory, hard drives, optical-disc drives, etc.
Computing device 100 includes one or more processors that read data
from various entities such as memory 112 or I/O components 120.
Presentation component(s) 116 present data indications to a user or
other device. Exemplary presentation components include a display
device, speaker, printing component, vibrating component, etc.
[0039] I/O ports 118 allow computing device 100 to be logically
coupled to other devices including I/O components 120, some of
which may be built in. Illustrative components include a
microphone, joystick, game pad, satellite dish, scanner, printer,
wireless device, etc.
[0040] Referring now to FIG. 2, a block diagram is provided
illustrating an exemplary system 200 in which embodiments of the
present invention may be employed. It should be understood that
this and other arrangements described herein are set forth only as
examples. Other arrangements and elements (e.g., machines,
interfaces, functions, orders, and groupings of functions, etc.)
can be used in addition to or instead of those shown, and some
elements may be omitted altogether. Further, many of the elements
described herein are functional entities that may be implemented as
discrete or distributed components or in conjunction with other
components, and in any suitable combination and location. Various
functions described herein as being performed by one or more
entities may be carried out by hardware, firmware, and/or software.
For instance, various functions may be carried out by a processor
executing instructions stored in memory.
[0041] Among other components not shown, the system 200 may include
a user device 202, a content server 204, a search engine 206, a
phrase generator 208, a phrase selection component 210, an
advertising delivery system 212, and storage 214. Each of the
components shown in FIG. 2 may be embodied on any type of computing
device, such as computing device 100 described with reference to
FIG. 1, for example. The components may communicate with each other
via a network 216, which may include, without limitation, one or
more local area networks (LANs) and/or wide area networks (WANs).
Such networking environments are commonplace in offices,
enterprise-wide computer networks, intranets, and the Internet. It
should be understood that any number of user devices, content
servers, search engines, phrase generators, phrase selection
components, advertiser delivery systems and storage may be employed
within the system 200 within the scope of the present invention.
Each may comprise a single device or multiple devices cooperating
in a distributed environment. For instance, the system 200 may
comprise multiple devices arranged in a distributed environment
that collectively provide the functionality of the system 200
described herein. Additionally, other components not shown may also
be included within the system 200.
[0042] In accordance with embodiments of the present invention, a
user may employ the user device 202 to submit search queries to the
search engine 206 and, in response, receive a search results page
with search results and advertisements. For instance, the user may
employ a web browser on the user device 202 to access a search
input web page and enter a search query. As another example, the
user may enter a search query via a search input box provided by a
search engine toolbar located, for instance, within a web browser,
the desktop of the user device 202, or other location. One skilled
in the art will recognize that a variety of other approaches may
also be employed for providing a search query within the scope of
embodiments of the present invention.
[0043] When the search engine 206 receives a search query from a
user device, such as the user device 202, the search engine 206
performs a search on a search system index 228, knowledge base 230,
and/or other data storage containing searchable content maintained
in storage 214. The search system index 228 may generally contain
unstructured and/or semi-structured data, while the knowledge base
230 may generally contain structured data. Accordingly, the search
engine 206 identifies a number of search results in response to the
received search query. Additionally, advertisements are selected
for inclusion in the search results page as will be described in
further detail below. In response to the search query, a search
results page may be provided to the user device 202 that includes
search results with selected advertisements.
[0044] The user may also employ the user device 202 to view web
pages hosted by content servers, such as web page 218 hosted on the
content server 204. For instance, the user may employ a web browser
on the user device 202 to request the web page 218 from content
server 204. In embodiments of the present invention, advertisements
are selected (as will be described in further detail below) for
inclusion on the web page 218 when presented to the user on the
user device 202.
[0045] The phrase generator 208, phrase selection component 210,
and advertising delivery system 212 generally operate to select and
deliver advertisements on search result pages and web pages. The
phrase generator 208 includes a lateral concept generator 220 and
semantic topic engine 222 that analyze a search query (and/or
associated search results) or a web page (and/or related content)
to identify phrases that may be used for advertisement selection.
The lateral concept generator 220 identifies lateral concepts for a
search query or web page, and the semantic topic engine 222
identifies topics for a search query or web page. In some
embodiments in the context of search, lateral concepts may be
included on a search results page to allow a user to employ the
lateral concepts to drill across a combination of structured,
unstructured, and semi-structured content. In some embodiments in
the context of search, a table of contents may be generated based
on identified topics, and the table of contents may be included on
a search results page. The table of contents lists identified
topics and allows the user to select topics and view search results
associated with each topic.
[0046] Lateral concepts and topics identified by the lateral
concept generator 220 and semantic topic engine 222 are treated as
candidate phrases for advertisement selection. The phrase selection
component 210 operates to analyze the candidate phrases and select
phrases that will be used to advertisement selection. In
embodiments, the phrase selection component 210 ranks the candidate
phrases, and the rankings are used for selecting phrases for
further processing. In some embodiments of the present invention,
candidate phrases are ranked based on monetization. In other words,
the candidate phrases are analyzed to estimate the extent to which
use of each candidate phrase to select advertisements will generate
advertising revenue. Candidate phrases that are determined to be
more likely to generate higher advertising revenue are ranked
higher. In various embodiments, one or more candidate phrases are
selected based on ranking. In some embodiments, the highest N
ranking candidate phrases (e.g., highest five ranking candidate
phrases) are selected. In other embodiments, candidate phrases
having a ranking greater than a threshold are selected. In further
embodiments, candidate phrases having a significantly higher
ranking than other candidate phrases are selected. Any and all such
variations of the above, as well as other considerations, may be
used to selected phrases from a group of candidate phrases.
[0047] The advertising delivery system 212 receives phrases from
the phrase selection component 210, selects advertisements, and
delivers the selected advertisements for presentation on a search
results page in response to a search query or for presentation on a
web page in response to an indication of the web page. The
advertisement delivery system 212 includes an advertisement
selection component 224 that queries the advertisement inventory
234 using selected phrases to select advertisements that will be
delivered for presentation to a user. The advertisement inventory
234 may store advertisements and metadata associated with each
advertisement. The metadata stored for an advertisement may include
information used in advertisement selection, such as for instance,
bid values from advertisers, click-through rates, etc. In
embodiments of the present invention, the advertisement selection
component 224 selects advertisements based on relevance of the
advertisements to selected phrases and/or based on monetization
(i.e., an estimate of the extent to which the advertisements will
generate advertising revenue). Auction processes currently employed
by advertising systems may be used for selecting advertisements.
Such processes are well known to those skilled in the art and
therefore will not be discussed in further detail herein.
[0048] The advertising delivery system also includes an
advertisement delivery engine 226 that facilitates delivery of
selected advertisement for presentation to a user. In the context
of search, a search results page is generated that includes search
results in response to the user's search query and the selected
advertisements. In the context of a web page, the advertisements
are delivered for inclusion in an area of the web page provided for
presentation of advertisements.
[0049] As indicated above, the lateral concept generator 220
generates lateral concepts in response to a search query and/or for
a web page. In one embodiment, such as that shown in FIG. 3, the
lateral concept generator 220 includes an initial processing
component 302, a similarity engine 304, and an indexing engine 306.
The lateral concept generator 220 receives categories and content
from storage 214. In turn, the content and categories are processed
by one or more components 302, 304, and 306 of the lateral concept
generator 220.
[0050] The initial processing component 302 is configured to locate
content that matches a search query or web page, to analyze the
content, and to extract information using one or more data
processing methods. In this regard, the initial processing
component 302 may be used to analyze content and extract
information from the three types of data: unstructured data,
structured data, and semi-structured data maintained by storage
214. Unstructured data may comprise documents with a series of text
lines. Documents that are included in the category of unstructured
data may have little or no metadata. Structured data, on the other
hand, may comprise a traditional database where information is
structured and referenced. Semi-structured data may comprise a
document such as a research paper or a Security and Exchange
Commission filing, where part of the document comprises lines of
text and part of the document comprises tables and graphs used for
illustration. In the case of semi-structured data, the structured
components of a document may be analyzed as structured data and the
unstructured components of the documents may be analyzed as
unstructured data.
[0051] Feature vectors are used to compare content matching the
search query or web page. The feature vectors may include the
following elements: a group of words, a concept, and score. The
group of words represents a summary or sampling of the content. The
concept categorizes the content, and the score contains a
similarity measure for the content and additional content matching
the search query or web page. For instance, a feature vector for
Space Needle content may include a group of words "monument built
for world fair" a concept "tower" and a score "null." The concepts
element of the feature vectors may be selected as the lateral
concept based on the score assigned to the feature vector.
[0052] The values for the elements of the feature vector may be
generated manually or automatically. A subject matter expert may
manually populate the elements of the feature vector.
Alternatively, the elements of the feature vector may be populated
automatically by the lateral concept generator 220.
[0053] The initial processing component 302 may include a lexical
analysis, a linguistic analysis, an entity extraction analysis, and
attribute extraction analysis. In an embodiment, the initial
processing component 302 creates feature vectors for the content in
storage 214. The initial processing component 302 automatically
populates the words and concepts for feature vectors. In certain
embodiments, the initial processing component 242 selects the
concepts from one or more ontologies in the ontology storage 232 of
storage 214, or from the words extracted from the content.
[0054] The similarity engine 304 calculates a similarity score that
populates the score element for the feature vector. The similarity
engine 304 is a component of the lateral concept generator 220. The
similarity engine 304 calculates a similarity score that is stored
in the feature vector for the content retrieved from storage 214.
The similarity score may represent similarity to other content in
storage 214 matching a received search query or identified web
page, similarity to the search query, or similarity to a web page,
such as web page 218. In turn, the similarity score is used to
select several categories from concepts identified in the feature
vectors associated with the content matching the search query or
web page. The selected categories are identified as lateral
concepts.
[0055] In one embodiment, the similarity engine 304 may calculate
similarity between content matching a search query or web page
using the feature vectors. The similarity score may be calculated
based on distance between the feature vectors using the Pythagorean
theorem for multidimensional vectors. For instance, when the
storage 214 includes content matching a received search query or
the web page 218, the lateral concept generator 220 may return
several categories based on scores assigned to content within each
of the several categories. The lateral concept generator 220
obtains the matching content and corresponding categories from
storage 214. In turn, the lateral concept generator 220 generates
the feature vector for the matching content. Also, the lateral
concept generator 220 generates a content collection using the
categories associated with the matching content. Each content in
the content collection is processed by the lateral concept
generator 220 to create feature vectors. In turn, each feature
vector for the content collection is compared to the feature vector
for the matching content to generate a similarity score. In turn,
the feature vectors for the content collection are updated with
similarity scores calculated by the similarity engine 302. The
similarity engine 302 may select a number of feature vectors with
high similarity scores in each category, average the scores, and
assign the category the averaged score. In an embodiment, the
similarity engine 302 selects three feature vectors within each
category assigned the highest score to calculate the average score
that is assigned to the categories. Thus, as an example, the top
five categories with the highest scores may be identified as
lateral concepts.
[0056] In another embodiment, the similarity engine 304 may
calculate similarity between content and a received search query or
a web page, such as web page 218. The similarity score may be
calculated based on distance between the feature vectors using the
Pythagorean theorem for multidimensional vectors. For instance,
when the storage 214 does not include content matching the search
query or the web page 218, the lateral concept generator 220 may
return several categories based on scores assigned to content
within each of the several categories. The lateral concept
generator 220 obtains a predetermined number of content related to
the search query or web page 218 and corresponding categories from
the storage 214. In one embodiment, the lateral concept generator
220 obtains a predetermined number (e.g., fifty) of content items
from storage 214 having a high similarity score for the search
query of web page 218. In turn, the lateral concept generator 220
generates a feature vector for the search query or web page 218.
Also, the lateral concept generator 220 retrieves a collection of
content using the categories associated with the obtained content.
Content in the collection of content is processed by the lateral
concept generator 220 to create feature vectors. In turn, the
feature vectors for content in the collection of content are
compared to the feature vector for the search query or web page 218
to generate a similarity score. In turn, the feature vectors for
the content collection are updated with similarity scores
calculated by the similarity engine 304. The similarity engine 304
may select a number of feature vectors with high similarity scores
in each category, average the scores, and assign the category the
averaged score. In an embodiment, the similarity engine 304 selects
three feature vectors within each category assigned the highest
score to calculate the average score that is assigned to the
categories. In turn, the top five categories with the highest
scores are identified as lateral concepts.
[0057] The similarity engine 304 may use word frequency to
calculate a query similarity score for the content in storage 214.
In one embodiment, the similarity engine 304 calculates a query
similarity score (S.sub.q) when a match to a search query is not
stored in the storage 214: S.sub.q= {square root over
(freq(w).times.log(docfreq(w)))}{square root over
(freq(w).times.log(docfreq(w)))}, where freq(w) is the frequency of
the query (w) in the storage and docfreq is the frequency of the
query within the content that is selected for comparison. The
content assigned the largest S.sub.q are collected by the
similarity engine 244, and the top fifty documents are used to
generate the lateral concepts.
[0058] The indexing engine 306 is an optional component of the
lateral concept generator 220. The indexing engine 306 receives the
lateral concepts from the similarity engine 304 and stores the
lateral concepts in storage 214 along with the search query or web
page that generates the lateral concepts. In turn, a subsequent
search query or web page request similar to a previously processed
search query or web page may bypass the lateral concept generator
220 and obtain the lateral concepts stored in the storage 214.
[0059] In embodiments, the storage 214 provides content and
previously generated lateral concepts. The storage 214 stores
content, ontologies, and advertisements. In certain embodiments,
the storage 214 includes one or more data stores, such as
relational and/or flat file databases and the like, that store a
subject, object, and predicate for each content. The storage 214
may reference content along with previously generated lateral
concepts. The content may include structured, semi-structured, and
unstructured data. In some embodiments, the content may include
video, audio, documents, tables, and images having attributes that
are stored in the flat file databases. The computer system 200 may
algorithmically generate the lateral concepts, or content
attributes may be used as lateral concepts.
[0060] For instance, content attributes for the Seattle Space
Needle or a particular stock may be stored in storage 214. The
content attributes may be provided as lateral concepts in response
to a search query for the Seattle Space Needle or the particular
stock, respectively. The Seattle Space Needle content attributes
may include a tower attribute, a Seattle attraction attribute, and
an architecture attribute. The tower attribute may include data
that specifies the name and height of the Seattle Space Needle and
other towers, such as Taipei 101, Empire State Building, Burj, and
Shanghai World Financial Center. The Seattle attraction attribute
may include data for the name and location of other attractions in
Seattle, such as Seattle Space Needle, Pike Place Market, Seattle
Art Museum, and Capitol Hill. The architecture attribute may
include data for the architecture type, modern, ancient, etc., for
each tower included in the tower attribute. Any of the Seattle
Space Needle content attributes may be returned as a lateral
concept.
[0061] A particular stock may include stock content attributes. For
instance, MSFT content attributes may include a type attribute, an
industry attribute, and a profit to earnings (PE) attribute. The
type attribute includes data for business type, e.g., corporation,
company, incorporated, etc. The industry attribute, may specify the
industry, e.g., food, entertainment, software, etc., and the PE
attribute includes the value of the PE. Any of the stock content
attributes may be returned as a lateral concept.
[0062] The lateral concepts that are generated algorithmically by
the lateral concept generator 220 may be stored in the storage 214.
In turn, subsequent search queries received by the search engine
220 or web page requests may be responded to, in certain
embodiments, with the lateral concepts stored in the storage 214.
For a given search query or web page, the storage 214 may store
several lateral concepts. Accordingly, the storage 214 may be
accessed to obtain a list of lateral concepts for a received search
query or requested web page. In some embodiments, the lateral
concepts may be provided as part of a search results page to enable
a user to navigate content in the storage 214.
[0063] The ontologies in the ontology storage 232 include words or
phrases that correspond to content in storage 214. The categories
associated with content in storage 214 may be selected from
multiple ontologies. Each ontology includes a taxonomy for a domain
and the relationship between words or phrases in the domain. The
taxonomy specifies the relationship between the words or phrases in
a domain. The domains may include medicine, art, computers, etc. In
turn, the categories associated with the content may be assigned a
score by the lateral concept generator 220 based on similarity. In
one embodiment, the lateral concept generator 220 calculates the
score based on similarity to content obtained in response to a
received search query or a requested web page. In another
embodiment, the lateral concept generator 220 calculates the score
based on similarity to the search query or web page. The lateral
concept generator 220 selects several categories as lateral
concepts based on the score.
[0064] The phrase generator 208 also includes a semantic topic
engine 222 that selects topics for received search queries or web
pages, such as web page 218. Topics identified by the semantic
topic engine 222 may be used to select advertisements for inclusion
on a search results page in response to a search query or for
inclusion on a requested web page, such as web page 218. In some
embodiments, topics identified by the semantic topic engine 222 for
a search query are used to generate a table of contents for
inclusion on a search results page that facilitates navigation of
the search results by a user. In such embodiments, in response to a
search query, a search results page may be provided to the user
device 202 that includes search results with a table of contents
that includes topics identified by the semantic topic engine
222.
[0065] As noted above, the semantic topic engine 222 identifies
topics relevant to received search queries or identified web pages.
As shown in FIG. 4, the semantic topic engine 222 generally
includes an ontology mapping component 402, an ontology topic
identification component 404, an ontology partial topic
identification component 406, a key-phrase topic identification
component 408, and a topic ranking/selection component 410. The
semantic topic engine 222 identifies semantic topics employing any
of components 402, 404, 406, and 408. In some embodiments of the
present invention, each of the components 402, 404, 406, and 408
may be employed to identify topics for a given search query or web
page, and the identified topics may be ranked and certain topics
selected by the ranking/selection component 410. In other
embodiments, topics may be identified by only a portion of the
components 402, 404, 406, and 408. For instance, in one embodiment,
once a threshold number of topics are identified by one or more of
components 402, 404, 406, and 408, further analysis by remaining
components is not performed. In further embodiments, the semantic
topic engine 222 may include only a portion of the components 402,
404, 406, and 408 shown in FIG. 4. Any and all such variations are
contemplated to be within the scope of embodiments of the present
invention.
[0066] When a search query is received or a web page is identified,
the ontology mapping component 402 operates to identify whether an
ontology mapping already exists for the search query or web page.
For instance, the search query may be a top-end search query for
which search system personnel have manually identified relevant
topics for a topics of contents for the search query. As another
example, the received search query or web page may correspond with
a search query or web page for which relevant topics have already
been identified and cached for the search query or web page. If the
ontology mapping component 402 determines that an ontology mapping
already exists for a received search query or web page, topics are
retrieved based on the ontology mapping. In some embodiments, only
topics retrieved by the ontology mapping component 402 or employed.
In other embodiments, additional topics are identified by one or
more of the other components 404, 406, and/or 408, as described in
further detail below.
[0067] The ontology topic identification component 404 operates on
a received search query or web page in conjunction with an ontology
of topics stored in the ontology storage 232 to identify relevant
topics for the search query or web page. The ontology storage 232
may store one or more ontologies, which are used by the ontology
topic identification component 404 to identify semantic concepts as
topics for received search queries or identified web pages. Each
ontology includes a collection of words and phrases defining
concepts and relationships between the concepts. In some
embodiments, a search is performed on the search system index 228,
knowledge base 230, and/or other content in the storage 214 to
retrieve search results for a received search query, and the
ontology topic identification component 404 analyzes the search
results in conjunction with the ontology of topics to identify
relevant topics for the search query. In further embodiments,
content in the storage 214 may be analyzed to identify content
relevant to a web page that may be used by the ontology topic
identification component 404 in conjunction with the ontology of
topics to identify relevant topics for the web page.
[0068] The ontology partial topic identification component 406
functions in a manner similar to the ontology topic identification
component 404 but uses an ontology of partial topics instead of an
ontology to topics. As used herein, a partial topic refers to
partially-named topics. Each partial topic includes a partial topic
identifier word that may be combined with an additional word or
phrase to create a topic. For example, "reviews" may be a partial
topic. When analyzed in context, the partial topic identifier word
"reviews" may be combined with additional words, such as, "expert"
or "user" to generate the topics "expert reviews" or "user
reviews." Accordingly, once partial topics are identified for a
search query or web page, the ontology partial topic identification
component 406 or an associated component names the partial
topic.
[0069] The key-phrase topic identification component 408 analyzes
search results for a received search query or content of a web page
(and possibly related content) to generate candidate key-phrases.
Generally, the key-phrase topic identification component 408
generates key-phrases from search results or content of a web page
and identifies independent key-phrases. The independent key-phrases
are evaluated to identify candidate topics.
[0070] A number of topics may be identified for a received search
query or identified web page by the ontology mapping component 402,
ontology topic identification component 404, ontology partial topic
identification component 406, and/or key-phrase topic
identification component 408. In some instances, all identified
topics may be considered as candidate phrases by the phrase
selection component 210. In other instances, a large number of
topics may be identified, and only a subset of the identified
topics is considered as candidate phrases by the phrase selection
component 210. In some embodiments, the semantic topic engine 222
includes a topic ranking/selection component 410 that operates to
rank and select topics as candidate phrases for further processing.
Topics may be ranked using a number of different factors in
accordance with various embodiments of the present invention. By
way of example only and not limitation, each topic may be ranked
based on the total number of documents assigned to each topic. A
larger number of documents assigned to a given topic may provide a
higher ranking for the topic. A topic may also be ranked based on
the ranking of each search result (or a selection of search
results--e.g., the top N search results) assigned to the topic. The
ranking of each search result corresponds with each search result's
relevance to the search query. Accordingly, more highly relevant
search results being assigned to a given topic may provide a higher
ranking for the topic. The length (e.g., number of words) of each
topic may further be used to rank the topics. Any and all such
variations are contemplated to be within the scope of embodiments
of the present invention. After ranking the candidate topics, the
topic ranking/selection component 410 selects phrases for further
processing.
[0071] Turning now to FIG. 5, a flow diagram is provided that
illustrates an overall method 500 for selecting advertisements for
placement on a search results page or web page in accordance with
an embodiment of the present invention. Initially, as shown at
block 502, a search query is received or a web page is identified
for which advertisements are to be selected and delivered for
presentation. Lateral concepts are identified for the search query
or web page by a lateral concept generator, such as the lateral
concept generator 220 of FIG. 2, as shown at block 504. Lateral
concepts may be identified, for instance, as described in further
detail below with reference to the method 600 of FIG. 6 and/or the
method 700 of FIG. 7. Relevant topics are also identified for the
search query or web page by a semantic topic engine, such as the
semantic topic engine 222 of FIG. 2, as shown at block 506. Topics
may be identified, for instance, as described in further detail
below with reference to the methods 800, 900, 1000, and 1100 of
FIGS. 8, 9, 10, and 11, respectively.
[0072] The lateral concepts and topics identified at blocks 504 and
506 are considered as candidate phrases for advertisement selection
at block 508. In particular, the candidate phrases from the lateral
concepts and topics are ranked and the rankings are used for
selecting phrases for further processing. In some embodiments of
the present invention, candidate phrases are ranked based on
monetization. In other words, the candidate phrases are analyzed to
estimate the extent to which use of each candidate phrase to select
advertisements will generate advertising revenue. Candidate phrases
that are determined to be more likely to generate higher
advertising revenue are ranked higher. In various embodiments, one
or more candidate phrases are selected based on ranking. In some
embodiments, the highest N ranking candidate phrases (e.g., highest
five ranking candidate phrases) are selected at block 508. In other
embodiments, candidate phrases having relevance greater than a
threshold are selected at block 508. In further embodiments,
candidate phrases having a significantly higher ranking than other
candidate phrases are selected. Any and all such variations of the
above, as well as other considerations, may be used to selected
phrases from a group of candidate phrases.
[0073] An advertisement inventory is queried using the selected
phrases at block 510 to select advertisements that will be
delivered for presentation to a user. In embodiments of the present
invention, the advertisements may be selected based on relevance of
the advertisements to selected phrases and/or based on monetization
(i.e., an estimate of the extent to which the advertisements will
generate advertising revenue). Auction processes currently employed
by advertising systems may be used for selecting advertisements.
Such processes are well known to those skilled in the art and
therefore will not be discussed in further detail herein.
[0074] As shown at block 512, the selected advertisements are
delivered for presentation to a user. In the context of search, a
search results page is generated that includes search results in
response to the user's search query and the selected
advertisements. In the context of a web page, the advertisements
are delivered for inclusion in an area of the web page provided for
presentation of advertisements.
[0075] A computer system may execute at least two
computer-implemented methods for dynamically generating lateral
concepts. In a first embodiment, the lateral concepts are selected
based on scores between feature vectors of content matching the
query and other content in storage. With reference to FIG. 6, a
flow diagram is provided that illustrates a method 600 for
generating lateral concepts in accordance with an embodiment of the
invention. The method initializes in block 602 when the computer
system is connected to a network of client devices. A search query
or identification of a web page is received at block 604. In turn,
the computer system obtains content that corresponds to the user
query or web page from storage, as shown at block 606. At block
608, the computer system identifies categories associated with the
obtained content corresponding with the received search query or
identified web page. In one embodiment, the categories include
phrases in one or more ontologies. In another embodiment, the
categories comprise attributes of the obtained content
corresponding with the received search query or identified web
page. In turn, the computer system retrieves, from storage, a
collection of content that corresponds to each identified category,
as shown at block 610.
[0076] At block 612, the computer system selects several identified
categories as lateral concepts based on scores assigned to content
in the collection of content. In one embodiment, the lateral
concepts may include orthogonal concepts. The lateral concepts may
be stored in the storage of the computer system.
[0077] In certain embodiments, the content is represented as
feature vectors, and the score is assigned to the content based on
similarity between feature vectors. The lateral concepts are used
in embodiments to identify phrases for selecting advertisements as
discussed hereinabove. In some embodiments, the lateral concepts
may also be provided to the user, for instance, in a search results
page with search results in response to a search. In such
embodiments, content displayed with the lateral concepts may be
filtered by the computer system based on the similarity score
assigned to the content.
[0078] The computer system may select, in some embodiments,
orthogonal concepts by identifying the normal to a plane
corresponding to the feature vector of the obtained content. In
turn, feature vectors for the collection of content that create
planes, which are parallel to a plane created by the normal, are
processed by the computer system to obtain categories of the
content associated with those feature vectors. Several of these
categories may be returned as lateral concepts based on a score
assigned to the content within the categories. The method
terminates at block 614.
[0079] As mentioned above, the computer system may execute at least
two computer-implemented methods for dynamically generating lateral
concepts. In a second embodiment, the lateral concepts are selected
based on scores between feature vectors for search query or web
page and content in storage. The computer system may execute this
method when the storage does not contain a match to the search
query or web page. In some embodiments, a match is determined
without using stems for the terms included in the search query or
web page. Thus, the storage of the computer system may include
other matches that are based on the stems of the terms included in
the search query or web page. These other matches may be used to
generate the lateral concepts.
[0080] Referring to FIG. 7, a flow diagram is provided that
illustrates an alternative computer-implemented method 700 for
generating knowledge content in accordance with an embodiment of
the invention. The method initializes at block 702 when the
computer system is connected to a network of client devices.
[0081] A search query or identification of a web page is received
at block 704. At block 706, the computer system calculates
similarity between content in storage and the search query or web
page. At block 708, the computer system creates a collection of
content having a predetermined number of content similar to the
search query or web page. In turn, the computer system identifies
each category that corresponds to content in the collection of
content, as shown at block 710. At block 712, the computer system
selects several identified categories as lateral concepts based on
scores assigned to content in the collection of content.
[0082] In certain embodiments, the search query or web page and
content are represented as feature vectors, and the score is
assigned to the content based on similarity between feature vectors
for the search query or web page and content. In embodiments, the
lateral concepts are used as phrases for advertisement selection as
discussed hereinabove. In some embodiments, lateral concepts may
also be to the user in a search results page in response to a
search query. In such embodiments, content displayed with the
lateral concepts may be filtered by the computer system based on
the similarity score assigned to the content.
[0083] In one embodiment, orthogonal concepts may be included in
the lateral concepts. The orthogonal concepts are selected by
identifying the normal to a plane corresponding to the feature
vector of the query. In turn, feature vectors for the collection of
content that create planes, which are parallel to a plane created
by the normal, are processed by the computer system to obtain
categories of the content associated with those feature vectors.
Several of these categories may be returned as lateral concepts
based on a score assigned to the content within the categories. The
method terminates at block 714.
[0084] Turning to FIG. 8, a flow diagram is provided that
illustrates a method 800 for identifying topics for a search query
or web page in accordance with an embodiment of the present
invention. As shown at block 802, a search query or identification
of a web page is received. In accordance with the embodiment shown
in FIG. 8, a determination is made at block 804 regarding whether
an ontology mapping already exists for the search query or web
page. For instance, the search query may be a top-end search query
for which search system personnel have manually identified relevant
topics for the search query. Alternatively, the received search
query or web page may correspond with a search query or web page
that has been previously processed to identify relevant topics, and
the system may have cached the identified topics for the search
query or web page. If it is determined that an ontology mapping
already exists at block 806, topics for the search query or web
page are retrieved at block 808. In some embodiments, only the
topics retrieved at block 808 are used as candidate phrases for
advertisement selection and the process ends. In other embodiments,
the process continues at block 810, and additional topics are
algorithmically identified.
[0085] If it is determined at block 806 that an ontology mapping
does not exist for the search query or web page (or if the process
continues after retrieving topics at block 808), the search query
or web page is further analyzed in conjunction with an ontology of
topics (or a collection of topic ontologies) to identify whether
any topics from the ontology are relevant to the search query or
web page, as shown at block 810. In embodiments in which a search
query is received, search results and/or other content identified
as relevant to the search query is compared to the ontology of
topics to identify relevant topics. In embodiments in which a web
page is identified, content of the web page and/or stored content
identified as being relevant to the web page is compared to the
ontology of topics to identify relevant topics.
[0086] Identifying a topic from the ontology of topics as being
relevant to a search query or web page may be performed in a number
of different manners within the scope of embodiments of the present
invention. By way of example only and not limitation, in one
embodiment, content associated with a search query or web page is
converted to one or more feature vectors based on words contained
in the content. Each feature vector is compared to topics in the
ontology to determine distance of the feature vector to the topics.
Positive topic identification may be based on determining that a
feature vector is within a predetermined distance of a given topic.
Analysis at block 810 may identify zero or more topics from the
ontology of topics as being relevant to the search query or web
page.
[0087] The search query or web page is compared to an ontology of
partial topics (or a collection of partial topic ontologies) at
block 812. As indicated above, a partial topic is a topic that is
only partially-named. Each partial topic includes a partial topic
identifier word that may be combined with an additional word or
phrase to create a topic.
[0088] Identifying a partial topic from the ontology of partial
topics as being relevant to a search query or web page may be
performed in a number of different manners within the scope of
embodiments of the present invention. By way of example only and
not limitation, in one embodiment, content associated with a search
query or web page is converted to one or more feature vectors based
on words contained in the content. Each feature vector is compared
to partial topics in the ontology to determine distance of the
feature vector to the partial topics. Positive partial topic
identification may be based on determining that a feature vector is
within a predetermined distance of a given partial topic. Analysis
at block 812 may identify zero or more partial topics from the
ontology of partial topics as being relevant to the search query or
web page.
[0089] Partial topics identified at block 812 are named, as shown
at block 814. A flow diagram is provided in FIG. 9 that illustrates
a method 900 for naming a partial topic in accordance with an
embodiment of the present invention. As shown at block 902,
occurrences of the partial topic identifier word within search
results and/or content associated with the search query (in the
case of a received search query) or content of the web page, and/or
content identified as being relevant to the web page (in the case
of an identified web page) are identified. For instance, the
partial topic identifier word may be "reviews," and each occurrence
of that term is identified. At block 904, one or more words and/or
phrases around the partial topic identifier word are extracted. The
frequency of each extracted word and/or phrase is counted, as shown
at block 906. In some embodiments, the location of each extracted
word and/or phrase with respect to the partial topic identifier
word is tracked and counted. In particular, a word or phrase may
appear before or after the partial topic identifier word. The
system may separately track how many times each word and/or phrase
appears before the partial topic identifier word and how many times
each word and/or phrase appears after the partial topic identifier
word.
[0090] After the search results and/or content have been analyzed,
the most frequently used word or phrase is selected, as shown at
block 908. Additionally, the partial topic is named using the
partial topic identifier word and the most frequently used word or
phrase, as shown at block 910. The sequencing of the partial topic
identifier word and the most frequently used word or phrase may be
determined based on the majority ordering in the analyzed text. For
instance, if the selected word or phrase occurred before the
partial topic identifier word more often than after the partial
topic identifier word, the sequence for the partial topic name will
include the selected word or phrase first followed by the partial
topic identifier word.
[0091] Returning to FIG. 8, topics are identified by extracting
key-phrases from search results and/or content relevant to a search
query or from the content of an identified web page and/or other
content relevant to the web page, as shown at block 816.
Identifying key-phrase topics may be performed by computing
independent key-phrases and selecting topics based on the
independent key-phrases. The process at block 816 may result in
zero or more key-phrase topics. Referring to FIG. 10, a flow
diagram is provided that illustrates a method 1000 for computing
independent key-phrases in accordance with an embodiment of the
present invention. As shown at block 1002, candidate key-phrases
are generated from search results and/or other content for a search
query or from the content of an identified web page and/or other
content relevant to the web page. In accordance with some
embodiments of the present invention, a Markov chain based method
is used to generate the candidate key-phrases.
[0092] Candidate key-phrases are evaluated for independence, as
shown at block 1004. Independence of candidate key-phrases may be
evaluated using a number of metrics in accordance with embodiments
of the present invention. For instance, independence may be
determined based on any combination of the following metrics: the
number of words shared between the candidate key-phrases, analysis
of acronyms of words in the key-phrases, and the number of
documents shared by candidate key-phrases.
[0093] For each group of mutually dependent key-phrases, the
mutually dependent key-phrases are merged at block 1006. As such,
the most frequent key-phrase from a group of mutually dependent
key-phrases is selected as a key-phrase for further analysis, as
shown at block 1008. The process of merging mutually dependent
key-phrases to identify key-phrases for further analysis is
repeated until no more mutually dependent key-phrases remain. The
result of the method 1000 is a collection of one or more
independent key-phrases that may be further evaluated as possible
topics.
[0094] Referring again to FIG. 8, a collection of candidate topics
is provided as a result of the above-described process and may
include topics identified from an existing ontology mapping,
analysis of an ontology of topics, analysis of an ontology of
partial topics, and/or key-phrase generation. In some instances, a
larger number of topics may have been identified than is desired.
As such, the process in some embodiments continues by ranking and
selecting topics for further analysis. As shown at block 818, the
candidate topics are ranked. Candidate topics may be ranked using a
number of different factors in accordance with various embodiments
of the present invention. By way of example only and not
limitation, each candidate topic may be ranked based on the
distance between features vectors to topics. A candidate topic may
also be ranked based on the total number of content items
identified as related to each candidate topic. A larger number of
documents assigned to a given candidate topic may provide a higher
ranking for the candidate topic. In the context of search, a
candidate topic may also be ranked based on the ranking of each
document (or a selection of documents--e.g., the top N documents)
assigned to the candidate topic. The ranking of each document
corresponds with each document's relevance to the search query.
Accordingly, more highly relevant documents being assigned to a
given candidate topic may provide a higher ranking for the
candidate topic. The length (e.g., number of words) of each
candidate topic may further be used to rank the candidate topics.
Any and all such variations are contemplated to be within the scope
of embodiments of the present invention.
[0095] As shown at block 820, topics are selected from the list of
candidate topics based on ranking. In some embodiments, a
predetermined number of topics is selected. For instance, the five
topics with the highest ranking may be selected. In other
embodiments, all topics having a ranking satisfying a predetermined
or dynamic threshold may be selected. In further embodiments,
topics having a significantly higher ranking than other topics are
selected. Any combination of the above and/or additional approaches
to selecting topics based on ranking may be employed within
embodiments of the present invention.
[0096] Topics selected via the method 800 may be further processed
to select advertisements as discussed hereinabove. Additionally, in
some embodiments, a table of contents may be generated based on the
selected topics. The table of contents may be included on a search
results page that is generated in response to the search query. For
instance, the table of contents may be presented in a side panel
adjacent to the search results or in another portion of the search
results page.
[0097] With reference now to FIG. 11, a flow diagram is provided
that illustrates a method 1100 for identifying topics for a search
query received at a search system in accordance with another
embodiment of the present invention. As shown at block 1102, a
search query is received. In accordance with the embodiment shown
in FIG. 11, a determination is made at block 1104 regarding whether
an ontology mapping already exists for the search query. For
instance, the search query may be a top-end search query for which
search system personnel have manually identified relevant topics
for the search query. Alternatively, the received search query may
correspond with a search query that has been previously processed
by the search system to identify relevant topics, and the search
system may have cached the identified topics for the search query.
If it is determined that an ontology mapping already exists at
block 1106, topics for the search query are retrieved at block
1108. In some embodiments, only the topics retrieved at block 1108
are used for advertisement selection and the process ends. In other
embodiments, the process continues at block 1110, and additional
topics are algorithmically identified.
[0098] If it is determined at block 1106 that an ontology mapping
does not exist for the search query (or if the process continues
after retrieving topics at block 1108), a search is performed using
the search query, as shown at block 1110. Search results for the
search query are returned, and the top N document snippets from the
search are received at block 1112 as a document set to be
analyzed.
[0099] As shown at block 1114, each document snippet in the
document set is compared to an ontology of topics (or a collection
of ontologies) to identify whether each document snippet maps to a
topic in the ontology. Identifying a document snippet as being
associated with a topic in the ontology of topics may be performed
in a number of different manners within the scope of embodiments of
the present invention. By way of example only and not limitation,
in one embodiment, a document snippet is converted to a feature
vector based on words contained in the document snippet, and the
feature vector is compared to topics in the ontology to determine
distance of the feature vector to the topics. Positive topic
identification is determined for a given document snippet by
determining that the feature vector for the document snippet is
within a predetermined distance of a given topic. If topic
identification is positive for a given document snippet at block
1116 based on analysis of the document snippet and ontology, the
document snippet is assigned to the identified topic, as shown at
block 1118. Additionally, the document snippet is removed from the
document set at block 1120.
[0100] After identifying a relevant topic for a given document
snippet (e.g., via blocks 1116-1120) or determining that no topic
from the ontology is sufficiently relevant for the document snippet
(e.g., via block 1116), a determination is made at block 1122
regarding whether the document snippet analyzed was the last
document snippet in the document set to be analyzed. If additional
document snippets remain for analysis, the process of blocks
1116-1122 is repeated until all document snippets in the document
set have been compared to the ontology of topics. After all
document snippets in the document set have been compared to the
ontology of topics, topics identified from the ontology of topics
are added to a list of candidate topics for consideration, as shown
at block 1124. In some embodiments, all identified topics are added
to the list. In other embodiments, only a portion of the topics are
added. For instance, in some embodiments, only topics having a
predetermined number of assigned document snippets are added to the
list of topics.
[0101] As shown at block 1126, each remaining document snippet in
the document set is compared to an ontology of partial topics (or a
collection of ontologies). As indicated above, a partial topic is a
topic that is only partially-named. Each partial topic includes a
partial topic identifier word that may be combined with an
additional word or phrase to create a topic.
[0102] Whether a given document snippet is associated with a
partial topic in the ontology of partial topics is determined at
block 1128. Identifying a document snippet as being associated with
a partial topic may be performed in a number of different manners
within the scope of embodiments of the present invention. By way of
example only and not limitation, in one embodiment, a document
snippet is converted to a feature vector based on words contained
in the document snippet, and the feature vector is compared to
partial topics in the ontology of partial topics to determine
distance of the feature vector to the partial topics. Positive
partial topic identification is determined for a given document
snippet by determining that the feature vector for the document
snippet is within a predetermined distance of a given partial
topic. If partial topic identification is positive for a given
document snippet at block 1128 based on analysis of the document
snippet and the ontology of partial topics, the document snippet is
assigned to the identified partial topic, as shown at block 1130.
Additionally, the document snippet is removed from the document set
at block 1132.
[0103] After identifying a relevant partial topic for a given
document snippet (e.g., via blocks 1128-1132) or determining that
no partial topic from the ontology is sufficiently relevant for a
given document snippet (e.g., via block 1128), a determination is
made at block 1134 regarding whether the document snippet analyzed
was the last document snippet in the document set to be analyzed.
If additional document snippets remain for analysis, the process of
blocks 1128-1134 is repeated until all document snippets in the
document set have been compared to the ontology of partial
topics.
[0104] After each document snippet remaining in the document set
has been compared against the ontology of partial topics, partial
topics are named at block 1136. In some embodiments, all identified
partial topics are named. In other embodiments, only a portion of
the topics are named and others are not considered for further
analysis. For instance, in some embodiments, only partial topics
having a predetermined number of assigned document snippets are
named and considered for further analysis. Partial topics may be
named as discussed above with reference to FIG. 9. Named partial
topics are added to the list of topics, as shown at block 1138.
[0105] Independent key-phrases are generated at block 1140 from the
document snippets remaining in the document set after comparison of
the documents snippets to the ontology of topics and the ontology
of partial topics. Independent key-phrases may be generated as
discussed above with reference to FIG. 10.
[0106] After identifying candidate topics from independent
key-phrases, document snippets remaining in the document set are
assigned to the key-phrase topics, as shown at block 1142.
Identifying a document snippet as being associated with a
key-phrase may be performed in a number of different manners within
the scope of embodiments of the present invention. By way of
example only and not limitation, in one embodiment, a document
snippet is converted to a feature vector based on words contained
in the document snippet, and the feature vector is compared to
key-phrases to determine distance of the feature vector to the
key-phrases. Positive key-phrase identification is determined for a
given document snippet by determining that the feature vector for
the document snippet is within a predetermined distance of a given
key-phrase. Key-phrase topics are identified as shown at block 1144
and added to the list of topics at block 1146. In some embodiments,
all independent key-phrases are identified as key-phrase topics and
added to the list of topics. In other embodiments, only a portion
of the key-phrases are recognized as topics are added to the list
of topics. For instance, in some embodiments, only key-phrases
having a predetermined number of assigned document snippets are
identified as key-phrase topics and added to the list of
topics.
[0107] A list of candidate topics is provided as a result of the
above-described process and may include topics identified from an
existing ontology mapping, analysis of an ontology of topics,
analysis of an ontology of partial topics, and/or key-phrase
generation. In some instances, a larger number of topics may have
been identified than is desired. As such, the process in some
embodiments continues by ranking and selecting topics for further
analysis. As shown at block 1148, the candidate topics are ranked.
Candidate topics may be ranked using a number of different factors
in accordance with various embodiments of the present invention. By
way of example only and not limitation, each candidate topic may be
ranked based on the total number of documents assigned to each
candidate topic. A larger number of documents assigned to a given
candidate topic may provide a higher ranking for the candidate
topic. A candidate topic may also be ranked based on the ranking of
each document (or a selection of documents--e.g., the top N
documents) assigned to the candidate topic. The ranking of each
document corresponds with each document's relevance to the search
query. Accordingly, more highly relevant documents being assigned
to a given candidate topic may provide a higher ranking for the
candidate topic. The length (e.g., number of words) of each
candidate topic may further be used to rank the candidate topics.
Any and all such variations are contemplated to be within the scope
of embodiments of the present invention.
[0108] As shown at block 1150, topics are selected from the list of
candidate topics based on ranking. In some embodiments, a
predetermined number of topics is selected. For instance, the five
topics with the highest ranking may be selected. In other
embodiments, all topics having a ranking satisfying a predetermined
or dynamic threshold may be selected. In further embodiments,
topics having a significantly higher ranking than other topics are
selected. Any combination of the above and/or additional approaches
to selecting topics based on ranking may be employed within
embodiments of the present invention.
[0109] Topics selected via the method 1100 may be further processed
to select advertisements as discussed hereinabove. Additionally, in
some embodiments, a table of contents may be generated based on the
selected topics. The table of contents may be included on a search
results page that is generated in response to the search query. For
instance, the table of contents may be presented in a side panel
adjacent to the search results or in another portion of the search
results page.
[0110] By way of illustration, FIGS. 12 and 13 include exemplary
screen displays showing presentation of advertisements selected in
accordance with embodiments of the present invention. It will be
understood and appreciated by those of ordinary skill in the art
that the screen displays of FIGS. 12 and 13 are provided by way of
example only and are not intended to limit the scope of the present
invention in any way.
[0111] Referring initially to FIG. 12, an exemplary screen display
is provided that shows a search results page 1200 including
advertisement selected in accordance with an embodiment of the
present invention. As shown in FIG. 12, the search results page
1200 has been provided in response to the search query 1202,
"sammamish fun." In response to the search query 1202, the search
results page 1200 includes a number of search results to the search
query 1202 that are provided in a search results area 1204.
Additionally, the search results page 1200 includes a left-side
pane providing a table of contents 1208 listing topics identified
for the search query and lateral concepts in a lateral concept area
1210 (specific lateral concepts have been omitted from the search
results page 1200). In the screen display of FIG. 12, "All Results"
are currently being displayed in the search results area 1204. If a
user selects a topic from the table of contents 1208, search
results relevant to the selected topic would be displayed in the
search results area 1204. Alternatively, if a user selects a
lateral concept from the lateral concept area 1210, content
relevant to the selected lateral concept would be displayed in the
search results area 1204. The search results page 1200 also
includes advertisements 1206. In accordance with embodiments of the
present invention, the advertisements have been selected using
phrases based on the topics and lateral concepts. As shown in FIG.
12, the search results page 1200 may include further features, such
as, for instance, related search queries 1212, and search history
1214. Details of these sections have been omitted from the search
results page 1200.
[0112] Turning to FIG. 13, an exemplary screen display is provided
that shows a web page 1300 including advertisement selected in
accordance with an embodiment of the present invention. As shown in
FIG. 13, the web page 1300 includes content 1302. In the present
example, the content 1302 is a news story regarding an injury to a
college football player. Note that the details of the news story
have been omitted from the web page 1300 in FIG. 13. The web page
1300 also includes a number of advertisements 1304 that have been
selected based on lateral concepts and topics identified from the
content 1302 in accordance with embodiments of the present
invention.
[0113] As can be understood, embodiments of the present invention
provide for the selection and delivery of advertisements on search
result pages and web pages based on lateral concepts and topics
identified based on analysis of search queries and web pages. The
present invention has been described in relation to particular
embodiments, which are intended in all respects to be illustrative
rather than restrictive. Alternative embodiments will become
apparent to those of ordinary skill in the art to which the present
invention pertains without departing from its scope.
[0114] From the foregoing, it will be seen that this invention is
one well adapted to attain all the ends and objects set forth
above, together with other advantages which are obvious and
inherent to the system and method. It will be understood that
certain features and subcombinations are of utility and may be
employed without reference to other features and subcombinations.
This is contemplated by and is within the scope of the claims.
* * * * *