Variable personalization of search results in a search engine

Description of equivalent WO 2005098611 A2

0001VARIABLE PERSONALIZATION OF SEARCH RESULTS IN A SEARCH ENGINE INVENTORS: Taher Haveliwala, Glen Jeh, Sepandar Ka var

CROSS REFERENCE TO RELATED APPLICATIONS

0003[0001] This application claims the benefit under 35 U.S.C. §119(e) of Application

0004Serial No. 60/557,783 filed on March 29, 2004, and Application Serial No. 11/002,474 filed on December 1, 2004 which are incorporated by reference herein. This application also incorporates by reference U.S. Patent Application Serial Nos. 10/646,331 (filed on 8/22/2003) and 10/641,489 (filed on 8/14/2003); all of the above applications are commonly owned along with this application. FIELD OF INVENTION

0005[0002] This invention relates in general to online search engines, and in particular to personalization of search results. BACKGROUND OF INVENTION

0006[0003] The current state of the art in online search engines is highly advanced in its ability to retrieve documents (e.g., web pages or portions thereof, images, etc.) that are responsive to the terms of a query. Search engines today can quickly retrieve specific documents that match the terms of the query. However, current search engines often return documents that while accurately correspond to the specific terms of the query, do not in any way reflect the user's underlying interests. Thus, two different users, one who is very interested in sports, and another who is interested in politics, will obtain exactly the same results to a given query, say "drug testing in baseball," even though the first user may be more interested in learning about which teams have implemented drug testing, while the latter user is more interested in learning about legislation related to drug testing in sports. Thus, there is a need for providing a mechanism and methodology for personalizing search results in accordance with the interests of the users.

0007[0004] Further, while a user may have particular set of interests that may be useful in processing their search query, a user may not want to always have such interests rnfluence the query results. Thus, it would be desirable to a provide a mechanism and a methodology by which the user can variably adjust the degree to which his interests influence the results of a given search query. SUMMARY OF THE INVENTION

0008[0005] The present invention provides allows a user to variably adjust the degree to which his or her personal interests influence th-e presentation of results of a given search query. The present invention has multiple embodiments, including a system for providing a search engine, a methodology for processing a search query, and a user interface for the user to variably adjust the degree of influence.

0009[0006] For a user's perspective, the present invention provides a mechanism for personalizing search results. The user defines a set of interests, for example as a set of topics or categories of interest to the user, as an interests profile. The user then submits a search query to a search engine. The search engine retrieves documents that are relevant to the query and presents these as search results. At first (though not necessarily) the search results are ranked without regard to the user's interests. This enables the user to see the relevance of the documents without influence ^"by his personal interests. The user can variably control ranking of the documents in t-he search result set. In a preferred embodiment, this is done by direct manipulation of graphical control element, a personalization control. This graphical element can have various graphical representations, and in one embodiment it is represented as a slider. Adjusting the control element adjusts the degree to which the results are ranked according to the user's interests, as expressed in his interest profile. In response to changes in the position of the control element, the search results are reranked, thereby allowing the user to see how his interests affect the ordering of the results.

0010[0007] In one embodiment, the ranking of search results operates to increase the relevance score of documents that are associated with the directory of topics. Each topic in the directory is associated with a group of documents (e.g., websites) which are deemed relevant to the topic. For each document (again, website), a boost value is provided which indicates the degree of relevance of the site to the topic. All pages within a given site get that site's boost value. The boost value may be understood as a scaling factor that represents the relevance of the ^ite to the topic During the query processing, the set of documents is returned, and scored without regard to their inclusion in the topic directory. In addition, a separate score for each document is determined in which the document's underlying relevance score is adjusted by its associated boost value if any, for example by multiplying the relevance score by the boost value, thereby scaling the relevance of the document to the query. Generally, only those documents which belong to sites listed in the directory will have a boosted relevance score, and other documents in the search results will not. The reranking of the documents in accordance with their boosted relevance scores is then a function of the position of the personalization control, either from no influence

0011(where the boost value has no effect on the relevance score) to complete influence, wherein boost value for each document is applied 100% to the relevance score, and any number of intermediate levels of application. The user can thus immedia-tely rerank the documents in the search results by adjusting the position of the control.

0012[0008] A system embodiment includes a topic directory for providing topics

0013(categories) to a user, a search engine and index for searching indexed websites in response to a user's query, and a personalization component for determ-ining for the search results at least one (and preferably a set) of rankings of the search results in accordance with the user's interests. A user interface embodiment includes a topical directory, with various topics, and graphical controls (e.g., checkboxes) in which the user cam select topics of interests, and a graphical control that allows the user to adjust the search result ranking in accordance with his interest profile.

0014[0009] The present invention is next described with respect to various figures and diagrams and technical information. The figures depict various embodiments of the present invention for purposes of illustration only. One skilled in the art will readily recognize from the following discussion that alternative embodiments of the illustrated and described structures, methods, and functions may be employed without departing from the principles of the invention.

BRIEF DESCRIPTION OF THE DRAWINGS

0016[0010] FIG. 1 is an entry web page for a search engine.

0017[0011] FIG. 2 is web page for editing a user's personal profile of topics.

0018[0012] FIG. 3 is another web page of topics, and subtopics.

0019[0013] FIG.4 is another web page of subtopics.

0020[0014] FIG. 5 shows the entry web page for the search- engine with the query

0021"Stanford".

0022[0015] FIG. 6 shows the initial results of the search on- query "Stanford", along with a control 608 for controlling a degree of influence of the user's Interest profile in ranking the search results.

0023[0016] FIG. 7 shows the ranking of the search results of FIG. 6 after the user has adjusted the control 608 to an intermediate position. [0017] FIG. 8 shows the ranking of the search results of FIG. 6 after the user has adjusted the control 608 to a maximum position.

0024[0018] FIG. 9 shows a web page of topics after deletion of a "Computer" topic and selection of a "Music" topic by the user.

0025[0019] FIG. 10 shows the ranking of the search results of FIG. 6 following the change in topic interests in FIG. 9. DETAILED DESCRIPTION

0026[0020] Referring now to FIG. 1 there is shown the entry (or landing page) 100 of a search engine. The search engine includes a query entry field 102 in which a u_ser enters the terms of a search query. A search button 106 submits the query to the search engine for processing. The first time the user accesses the search engine, he will not have defined a user profile for personalizing the search results, and a message is displayed to that effect. The user can access the Edit Profile link 104 in order to establish their personal interests. [0021] Clicking on the Edit Profile link 104 accesses the page 200 illustrated in FIG. 2.

0027This page shows the top level topics 204 of a topical directory 202. The topics in the directory can be, for example, the topics defined in the Open Directory Project,, found at http: / / dmoz.org/ or can be some other topical directory. (In this disclosure t-he term "topic" and "category" are interchangeable, as are "subtopic" and "subcategorry".) The top level topic links may have a set of subtopics, as indicated by the arrows 206. At any time, the user can begin a search by clicking on the Start Searching button 208. Clicking on a particular top level topic takes the user to a page associated with that topic's s-ubtopics. [0022] FIG. 3 illustrates the result of this step, where the user has clicked on the topic

0028"Computers" on page 200. As shown in FIG. 3, the subtopics 302 of the topic "Computers" have been displayed in the right hand column. Next to each subtopic 302 is a checkbox 314. The user can select the checkbox 314 to indicate his interest in a specific subtopic. Alternatively, the user can select the checkbox 316 for the entire subtopic itself, "Computers (General)". This indicates that the user is interested in all of the subtopics of the topic "Computers". The set of topics selected by the user may be called a user profile, a personalization profile, or an interest profile. On the left side of FIG. 3, after the user has selected one or more topics of interest, these topics are listed in the interests field 306. Next to each topic name is a delete link 308 that allows the user to delete individual topics. The user can also delete all topics by selecting the "Delete all preferences" link 310. [0023] The subtopics themselves can have any number of additional levels of subtopics. For example, FIG.4 illustrates that the subtopic "Games" has furthe-r subtopics 402, "Board Games," "Card Games," etc. These were presented to the user in response to the user clicking on the subtopic arrow 206 next to the "Games" topic label. Again, the user can select either the entire subcategory by selecting checkbox 404, or any of the individual subtopic checkboxes 406. In one embodiment, in addition to the top level topics 204, there are two subtopic levels. This provides sufficient breadth and granularity of topical coverage, without imposing unnecessary computational complexity. [0024] Referring again to FIG. 3, the user has indicated his interest in computers generally. Clicking on the Start Searching button 208 brings up search page 10O again, which no longer shows the message requesting the user to establish a personalization profile. At this point, the user enters a search query in the search field, here the query "Stanford", as shown in FIG. 5. The search engine performs a search of its inderx and provides the search results. FIG. 6 illustrates the initial search results page 600. Here the search results 602 comprise a set of documents 604, each document indicated by a URL 617, along with a link 615 to the page, and a snippet 619 of the page text to show a portion of the context to the document to the user. Near the top of the page is the personalization control 608. In the context of this application, "documents" are understood to be any type of media that can be indexed and retrieved by a search engine, including web pages, images, multimedia files, text documents, PDFs, and so forth. No limitation as to the scope of the invention is implied by the use of the generic term "documents."

0029[0025] The personalization control 608 allows the user to adjust the ranJking of the individual documents/ pages of the search results with respect to his interests, preferably as expressed in the personalization profile. The personalization control may be explained as controlling a degree of influence or contribution of the user's interest profile to the ranking of the search results. FIG. 6 illustrates the control 608 in an embodiment where it is a slider type control that includes a control handle 610 that is moved along a linear dimension between a minimum value 612 to a maximum value 614. This type of control 608 is generally very intuitive and easy for users to understand how it operates. Alternative embodiments include dial type controls, similarly providing a variable adjustment between a minimum and maximum value, and radio buttons each with an associated predetermined adjustment amount (e.g., five radio buttons, associated respectively with 0%, 20%, 40%, 80%, and 100%). Those of skill in the art of user interface design will appreciate that other types of controls may be implemented to achieve essentially the same effect. [0026] In the example of FIG. 6, the personalization control 608 is set all the way to its minimum value 612. The search results accordingly are ranked on their relevance to the query and query independent information such as PageRank, as disclosed in U.S. Patent NJo. 6,285,999, "Method for node ranking in a linked database", which is incorporated by reference herein in its entirety and for all of its teachings. Notice that the search results are generally related to Stanford University.

0030[0027] Referring now to FIG. 7, there is shown the ranking of the search results 602 after the user has adjusted the slider 610 to an intermediate position about halfway between the minimum and maximum values. Now, the individual documents 604 are reranked, a-nd those that are more related to the user's interest (previously identified as "Computers (General)" are more highly ranked. In order to for the user to distinguish the documents that, are more related to his interests, those documents are indicated by a personalization icon 618 next to the link text 615. Notice that with the intermediate position of the slider 610, about every other document is a personalization result, though this is not required. Notice further that the snippet 619 text has text that relates to computers (e.g., "hacking") even though that term was not included in the original search query.

0031[0028] Referring now to FIG. 8, there is shown the ranking of the search results 60-2 after the user has adjusted the slider 610 to its maximum position 614. First, in this example all of the documents 604 in the search results 602 reflect the user's personalization profile_^ and are annotated with the personalization icon 618. Second, the documents in the search--, results relate to computers and Stanford University, thereby reflecting the user's interests in computers, as it relates to the query "Stanford". For example, the first two results are rela-ted to "hacking" at Stanford; the next three documents relate specifically to various computer- science projects at Stanford. Thus the third document is for Stanford's Artificial Intelligen-ce Laboratory, the fourth document is for Stanford's Center for the Study of Language and Information, and the fifth document is for Stanford's Database Group. The remaining documents are likewise related to the "Computers" interest of the user. Notice further, ti at what was previously the first (e.g., most relevant) result, the link to Stanford University's home page, when there was no personalization applied (FIG. 6) and when only intermediate personalization was applied (FIG. 7), no longer even appears in the top eight results. [0029] In the examples of FIGS. 6-8 the user can dynamically adjust the ranking of the documents in real time, without the search engine having to recompute the search results. This further aids the user's ability to explore the search results. [0030] As shown on FIGS. 6-8, an Edit Profile link 616 allows the user to edit the interests in his user profile at any time. FIG. 9 illustrates where the user has returned to the topic directory page 200 as illustrated in FIG. 2, and deleted his existing topic interest in Computers, and created a different interest in "Music". The user then again selects the Start Searching button 208. The search engine provides an updated set of search results, which doctunents will be the same as before, since the query term has not changed. However, as o-f the last search, the user has positioned the control 610 for full personalization, and hence the search engine applies this setting and ranks the documents according the existing (and new^ personalization profile, this time for "Music." FIG. 10 illustrates the dramatic difference this makes in the search results. Here, the top six results are all related to music and arts, and none of the previous results relating to computers make the top of the search result list. This shows how the user's personalization profile can significantly alter the ranking of a given set of documents.

0032[0031] In one embodiment of the invention, the personalization of search results is provided by the following methodology. Those of skill in the art of search engine design can readily understanding the basic structure of a search engine and its operation in which this methodology can be implemented.

0033[0032] 1) When the user selects topics (categories), they are added to a cookie containing a list of his interests IN (where IN is short for "interests"). The interests can be expressed as either text tokens (e.g. "computers"), or as indices into a directory, where each topic/ subtopic has an index number. When the user accesses the site, the system retrieves this interest list from the cookie. Alternatively, the site can provide a user registration page_.-, in which the user registers with a user name and password, and the user's interests are stored at the search engine site in a database, and retrieved in response to the user login. [0033] 2) Each time the user issues a search query Q, the list IN is sent to a frontend web server.

0034[0034] 3) Then the frontend web server issues two requests to a backend search engine (Q, null) and (Q, IN), which returns to the frontend web server two results sets, R_b^ase (normal) and R_in (personalized), respectively. [0035] 4) The base query (Q, null) is the same as searching for Q on the search engine without any personalization, so RJbase is the result set that the search engine would normally return in response to the query. As noted above, the search results accordingly are ranked on their relevance to the query and any other factors, such as PageRank, as disclosed in U.S. Patent No. 6,285,999, "Method for node ranking in a linked database", which is incorporated by reference herein in its entirety and for all of its teachings.

0035[0036] 5) The personalized query (Q, IN) contains the user's interest list. The search engine follows a similar processing control flow as a normal query, but it activates a personalization component during the scoring function, which "boosts" documents that are relevant to the set of interests IN.

0036[0037] In one embodiment of the invention, these boosts are implemented using a set of maps (one map for each topic/ subtopic in the directory), of the form:

0037[0038] {Site --> boost}

0038[0039] So for example, the category "Health" has a map of the form

0039[0040] {Site -> boost}

0040[0041] which contains entries such as:

0041[0042] nih.gov --> 5.8

0042[0043] cdc.gov -> 7.9

0043[0044] med.stanford.edu -> 3.5

0044[0045] where nih.gov, cdc.gov, and med.Stanford.edu are various sites that have been determined (either manually or automatically) to be related to the topic "Health". In other words, for each topic in the directory, there is a set of sites that have been determined to be relevant to the topic, and for each of these sites, a boost value is defined.

0045[0046] The boost for the sites listed in the topical directory is generally determined as follows:

0046[0047] a) A "site graph" is generated where nodes of the graph are sites (basically, pages on the same host) and edges between nodes are weighted based on the number of pages from one site that link to pages on another. This same type of graph can be used to compute all topic boost maps.

0047[0048] b) For each topic in the directory, say "Health", a number of sites are selected as "start sites" SO whose home page is listed in the Open Directory. For example, for a university like Stanford, start sites may be selected as any site ending in .stanford.edu.

0048[0049] c) A computation is run in two passes: [0050] i) first identify a set of sites SI that are linked-to heavily by those sites in SO, with each site in SI assigned a weight according to how heavily it's linked-to by sites in SO.

0049[0051] ii) then identify those sites S2 that are linked-to heavily by those sites in SI, weighted as with SI.

0050[0052] d) The sites in S2 are boosted with their assigned weights.

0051[0053] When a personalized query comes in with an interest set (e.g., {Health,

0052Music}) , then the personalization component consults the respective maps, and then for query results (i.e. documents) that are in these maps, the personalization components boosts the IR score for them, so that they would get ranked higher. The IR score is the information retrieval score that the search engine provides for each document, and can be a function of document relevance and other factors, such as PageRank.

0053[0054] For example if http://www.med.stanford.edu/research/ were one of the results for query Q, and IN included "Health," then that result would get a multiplicative boost of 3.5 to its IR score, because that page is on the site med.stanford.edu. For multiple interests, the values in the corresponding maps are multiplied together for a combined effect. Further details of one implementation for determining the boost for sites in the directory boost map are described in U.S. Patent Application Serial No. 10/646,331 incorporated by reference herein.

0054[0055] 6) The frontend web server receives RJbase and R_in. RJbase represents the results 602 the user sees when the personalization control 608 is set at 0 (or its minimum position 612). Rjn represents the results 602 the user sees when the slider is at its maximum position 614, e.g. 10. In one embodiment, the control has 11 positions, but any number of positions may be used. The rankings that are shown to the user between 0 and 10 are interpolated; this means using RJbase and R_in, the personalization component constructs intermediate rankings (R_l ... R_9) by moving each individual result along a virtual path from its position when the control 608 is at its minimum position to when the control 608 is at its maximum position.

0055[0056] 7) How the slider effect is provided to the user:

0056[0057] The frontend web server appends the results in R_in that are not in R_base to

0057RJbase to get a list of all unique results (either in RJbase or RjLn) called R_all. The frontend web server then assigns consecutive identification numbers to the results in R_all that serve as identifiers. The frontend then sends to the user's browser two items: [0058] a) the mapping {id -> result} for each of the results in R_all, containing for each document in the results its identification number; and

0058[0059] b) N vectors that denote the rankings the user should see, in terms of result

0059IDs, where N is the number of positions of the slider, and each vector represents the ordering of the results based on the degree of personalization. In this example, there would be 11 vectors corresponding to positions 0 through 10 of the control 608. In one embodiment Vector 0 is [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10], reflecting the un-personalized ranking of the results, assigned with the local IDs starting with R_0. Vector 10 is the ordering in the fully personalized results R_in; for instance, it might be [15, 17, 5, 0, 1, 2, 3, 4, 8, 9, 10], where documents 15 and 17, for example, are documents in the relevant map of the topic directory whose IR score was boosted according to the boost values in the map, and the value associated with the control 608's position. Transmitting the N ranking vectors to the client allows the client browser to quickly recomputed the rankings, rather than imposing that calculation on the server.

0060[0060] 8) Finally, Javascript code in the result page causes the user's browser to reorder the search results using the appropriate one of the 11 positional ranking vectors in response to the user changing the position of the control 608. Further details of one embodiment of steps 6-8, above, are described in U.S. Application Serial No. 10/641,489, incorporated by reference herein.

0061[0061] The present invention has been described in particular detail with respect to one possible embodiment. Those of skill in the art will appreciate that the invention may be practiced in other embodiments. First, the particular naming of the components, capitalization of terms, the attributes, data structures, or any other progr-unming or structural aspect is not mandatory or significant, and the mechanisms that implement the invention or its features may have different names, formats, or protocols. Further, the system may be implemented via a combination of hardware and software, as described, or entirely in hardware elements. Also, the particular division of functionality between the various system components described herein is merely exemplary, and not mandatory; functions performed by a single system component may instead be performed by multiple components, and functions performed by multiple components may instead be performed by a single component.

0062[0062] Some portions of above description present the features of the present invention in terms of algorithms and symbolic representations of operations on information. These algorithmic descriptions and representations are the means used by those skilled in the data processing arts to most effectively convey the substance of their work to others skilled in the art. These operations, while described functionally or logically, are understood to be implemented by computer programs. Furthermore, it has also proven convenient at times, to refer to these arrangements of operations as modules or by functional names, without loss of generality.

0063[0063] Unless specifically stated otherwise as apparent from the above discussion, it is appreciated that throughout the description, discussions utilizing terms such as "calculating" or "determining" or "identifying" or the like, refer to the action and processes of a computer system, or similar electronic computing device, that manipulates and transf orms data represented as physical (electronic) quantities within the computer system memories or registers or other such information storage, transmission or display devices. [0064] Certain aspects of the present invention include process steps and instructions described herein in the form of an algorithm. It should be noted that the process steps and instructions of the present invention could be embodied in software, firmware or hardware, and when embodied in software, could be downloaded to reside on and be operated from different platforms used by real time network operating systems. [0065] The present invention also relates to an apparatus for performing the operations herein. This apparatus may be specially constructed for the required purposes, or it may comprise a general-purpose computer selectively activated or reconfigured by a computer program stored on a computer readable medium that can be accessed by the computer. Such a computer program may be stored in a computer readable storage medium, such as, but is not limited to, any type of disk including floppy disks, optical disks, CD-ROMs, magnetic-optical disks, read-only memories (ROMs), random access memories (RAMs), EPROMs, EEPROMs, magnetic or optical cards, or any type of media suitable for storing electronic instructions, and each coupled to a computer system bus. Those of skill in the art of integrated circuit design and video codecs appreciate that the invention can be readily fabricated in various types of integrated circuits based on the above functional and structur d, descriptions, including application specific integrated circuits (ASIC). In addition, the present invention may be incorporated into various types of video coding devices.

0064[0066] The algorithms and operations presented herein are not inherently related to any particular computer or other apparatus. Various general-purpose systems may also be used with programs in accordance with the teachings herein, or it may prove convenient to construct more specialized apparatus to perform the required method steps. The required structure for a variety of these systems will be apparent to those of skill in the art, along with equivalent variations. In addition, the present invention is not described with reference to any particular programming language. It is appreciated that a variety of programming languages may be used to implement the teachings of the present invention as described herein, and any references to specific languages are provided for disclosure of enablement and best mode of the present invention.

0065[0067] Finally, it should be noted that the language used in the specification has been principally selected for readability and instructional purposes, and may not have been selected to delineate or circumscribe the inventive subject matter. Accordingly, the disclosure of the present invention is intended to be illustrative, but not limiting, of the scope of the invention.