CA2513851A1 - Phrase-based generation of document descriptions - Google Patents

Abstract

An information retrieval system uses phrases to index, retrieve, organize a nd describe documents. Phrases are identified that predict the presence of othe r phrases in documents. Documents are the indexed according to their included phrases. Related phrases and phrase extensions are also identified. Phrases in a query are identified and used to retrieve and rank documents. Phrases are also used to cluster documents in the search results, create document descriptions, and eliminate duplicate documents from the search results, and from the index.

Description

PHRASE-BASED GENERATION OF DOCU.MEN'I' DESCRIPTIONS
Inventor: Arena L. Patberson Crbss Reference to Related Applications [0001 ~ The application is related to the following co-pending applications:
Phrase Identification in an Information Retrieval System, Application No.10/
xxxxxx, filed o.~ July 26, 2004;
PhraseLBased Indexing in an Information Retrieval System, Application No.10/xxxxxx, filed on July 26, 2~4;
Phrasei-Based Searching in an Information Retrieval System, Application No.
20/xxxxxx, fled on July 26, 2004;
Phrase'-Based Personalization of Searches in an Information Retrieval System, Application No.10/xxxxxx, filed on July 26, 2004;
Anton~atic Taxonomy Generation in Search Results Using Phrases, Application No.
10/xx~,xacx, filed on July 26, 2004; and Phrase-Based Detection of Duplicate Documents in an information Retrieval System, Application No. 20/ xxxxxx, filed on July 26, 2004, all of which are co-owned, and incorporated by reference herein.
Field of the Invention [0002j; The present invention relates to an information retrieval system fvr indexi.~g, searching; and classifying documents in a large scale rnrpus, such as the Internet.
Background of the Invention [0003] Information retrieval systems, generally called search engines, are now an essent{aI tool for finding information in large scale, diverse, and growing corpuses such as the Internex Generally, search engines create an index that relates documents {or .'pages.') to the individual words present in each document. A document is retrieved in response to a query containing a number of query terms, typically based on having some riumber of query terms present in the document. The retrieved documex~ts are then ranked according to other statistical measures, such as frequency of occurrence of the query terms, host domain, link analysis, and the like. The retrieved documents are then ptesented to the user, typically in their ranked order, and without any further grouping or imposed hierarchy. In some cases, a selected portion of a text of a documrnt is presented to provide the user with a glimpse of the document's content.
~0004)I Direct "Boolean" matching of query terms has well known linnitations, and inparticular does not identify documents that do not have the query terms, but have r~iated words. For example, in a typical Boolean system, a search on "Australian Shephlards" would not return documents about other herding dogs such as Border Collies that do not have the exact guery terms. Rather, such a system is likely 6o also retrieve and highly rank documents that are about Australia (and have nothing to do with dogs), and documents about "shepherds" generally.
10405] The problem here is that conventional systems index documents based on individual terms, than on concepts. Concepts are often expressed in phrases, such as "Ausi~alian Shepherd," "President of the United States," or "Sundance F~7m Festival".
At best some prior systems will index documents with respect to a predetermined and very limited set of 'known' phrases, which are typically selected by a human operator.
Indexing of phrases is typically avoided because of the perceived computational and memory requirements to identify alI possible phrases of say three, four, ox five or more words: For example, on the assumption that any five words could constitute a phrase, and a Iarge corpus would have at least 20(3,U00 unique terms, there would approximately 3.2 x 1t?~ possible phrases, clearly more than any existing system could

2 store i~ memory or otherwise programmatically manipulate. A further problem is that phrased continually enter and leave the lexicon in terms of their usage, much more frequefitIy than new individual words are invented. New phrases are always being genera fed, from sources such technology, arts, world events, and law. Uther phrases will ae~ctine in usage over time.
(OOOf] Some existing information retrieval systems attempt to provide retrieval of concepts by using co-occurrence patterns of individual words. In these systems a search pn one word, such as "President" will also retrieve documents that have other words (that frequently appear with "President", such as "White" and " House."
While this approach may produce search results having documents that are conceptually related; at the level of individual words, it does not typically capture topical relationships that inhere between co-occurring phrases.
(0007] Accordingly, there is a need for an information retrieval system and methodology that can comprehensively identify phrases in a large scale corpus, index documents according to phrases, search and rank documents in accordance with their phrase, and provide additional clustering and descriptive information about the documents.
Sur~marX,of the Invention [0008] An information retrieval system and methodology uses phrases to index, search rank, and describe documents in the document collection. The system is adapted to identify phrases that have sufficiently frequent andJ or distinguished usage in the document collection to indicate that they are "valid" or "good"
phrases. In this manner multiple word phrases, for example phrases of four, five, or more terms, can be

3 identified, This avoids the problem of having to identify and index every possible phrasefs resulting from the all of the possible sequences of a given number of words.
~0009~I The system is further adapted to identify phrases that are related to each other, based on a phrase's ability to predict the presence of other phrases in a document.
More ~peci~cally, a prediction measure is used that relates the actual co-occurrence rate of twolplu'ases to an expected co-occurrence rate of the two phrases.
Information gain, as the ratio of actual co-occurrence rate to expected co-~currence rate, is one such prediction measure. Two phrases are related where the prediction measure exceeds a predei~rmined threshold. In that case, the second phrase has significant information gain with respect to the first phrase. Semantically, related phrases will be those that are commpnly used to discuss or describe a given topic or concept, such as "President of the United States" and "White House." For a given phrase, the related phrases can be ordered according to their relevance or significance based on their respective prediction measu.~es.
~0010~ An information retrieval system indexes documents in the document collection by the valid or good phrases. For each phrase, a posting list identifies the documlents that cor<tain the phrase. In addition, for a given phrase, a second list, vector, or other structure is used to store data indicating which of the related phrases of the given ~bhrase are also present in each document containing the give phrase. In this mamtelr, the system can readily identify not only which documents contain which phrased in response to a search query, but which documents also contain phrases that are related to query phrases, and thus more likely to be specifically about the topics or rnncerits expressed in the query phrases.

4 [fl0llj The use of phrases and related phrases further provides for the creation and use of clusters of related phrases, which represent semantically meaningful groupsk~gs of phrases. Clusters are identified frown related phra~s that have very high predm aon measure between all of the phrases in the cluster. Qusters can be used to organise the results of a search, including selecting which documents to include in the search;results and their order, as well as eliminating documents from the search results.
j0012j The information retrieval system is also adapted to use the phrases when searching far documents in response to a query. The query is processed to identify any phrased that are present in the query, so as to retrieve the associated posting lists for the query phrases, and the related phrase information. In addition, in some instances a user may enter an incomplete phrase in a search query, such as "President of the'.
Incomplete phrases such as these may be identified and replaced by a phrase extension, such as "President of the United States." This helps ensure that the user's most likely search ps in fact executed.
[U013] ~ The related phrase information may also be used by the system to identify or select which documents to include in the search result. The related phrase informlation indicates for a given phrase and a given document, which related phrases of the gi~~ea phrase are present in the given document. Accordingly, for a query containing two q~iQry phrases, the posting list for a first query phrase is processed to identify documents containing the first query phrase, and then the related phrase information is processed to identify which of these documents also contain the second query phrase.
These latter documents are then included in the search results. This eliminates the need for the; system to then separafiely process the posting list of the second query phrase, thereto providing faster search times. ~f course, this approach may be extended to any number of phrases in a query, yielding in significant computational and timing savings.
[0014 The system may be further adapted to use the phrase and related phrase information to rank documents in a set of search results. The related phiase information of a given phrase is preferably stored in a format, such as a bit vector, which expre~es the relative significance of each related phrase to the given phrase.
For example, a related phrase bit vector has a bit for each related phrase of the given phrase, and this bits are oxdered according to the pzediction measures (e.g., infomnation gain) for the related phrases. The mast significant bit of the related phrase bit vector are associ ted with the related phrase having the highest prediction measure, and the least significant bit is associated with the related phrase having a lowest prediction measure.
In this~manner, for a given document and a given phrase, the related phrase information can be; used to score the document. The value of the bit vector itself (as a value) may be used as the document score. In this manner documents that contain high order related phrases of a query phrase are more likely to be topically related to the query than those that hive low ordered related phrases. The bit vector value may also be used as a compqnent in a more complex scoring functior~, and additionally may be weighted. The documents can then be ranked acco~zding to their document scores.
(0015]I Phrase inforrnatian may also be used in an information retrieval system to personalize searches for a user. A user is modeled as a collection of phrases, for example, derived from documents that the user has accessed (e.g., viewed on screen, printed, stored, etc.). More particularly, given a document accessed by user, the related phrases that are present in this document, are included in a user model or profile.

Duryn~ subsequent searches, the phrases in the user model are used to filter the phrases of the search query and to weight the document scores of the retrieved documents.
[001fij Phrase information may also be. used in an information retrieval system to creaxe a description of a document, for example the documents included in a set of searchiresults. Given a search query, the system identifies the phrases present in the query,along with their related phrases, and their phrase extensions. For a given document, each sentence of the document has a count of how many of the query phrases, related phrases, and phrase extensions are present in the sentence.
The sentences of document can be ranked by these counts ('individually or in combination), and solme number of the top ranking sentences (e.g., five sentences) are selected to form the document description. The document description can then be presented to the user when the document is included in search results, so that the user obtains a better understanding of the document, relative to the query.
[0017] A further refinement of this process of generating document descriptions allows ahe system to provide personalized descriptions, that reflect the interests of the user. Ass before, a user model stores information identifying related phrases that are of interest to the user. This user model is intersected with a list of phrases related to the query phrases, to identify phrases common to both groups. The cowman set is then ordered according to the related phrase information. The resulting set of related phrases is then;used to rank the sentences of a document according to the number of instances of these rated phrases present in each document. A number of sentences having the highest number of common related phrases is selected as the personalized document description.

[Ob18] An information retrieval system may also use the phrase information 6o identify and eliminate duplicate documents, either while indexing (crawling) the documQnt collection, or when processing a search query. For a given document, each sentence of the document has a count of how many related phrases are present in the sentence. The sentences of document can be ranked by this count, and a number of the top rarllCing sentences (e.g., five sentences) are selected to form a document description.
This description is then stored in association with the document, for example as a string or a hash of the sentences. During indexing, a newly crawled document is processed in the safe manner to generate the document description. The new document description can be matched (e. g., hashed) against previous document descriptions, and if a match is found, then the new document is a duplicate. Similarly, during preparation of the results ~f a search query, the documents in the search result set can be processed to eliminate duplicates.
[00191 The present invention has further embodiments in system and software architectures, computer program products and computer implemented methods, arid computer generated user interfaces and presentations.
0020] The foregoing are just some of the features of an information retrieval system: and methodology based on phrases, Those of skill in the art of information retrieval will appreciate the flexibility of generality of the phrase information allows for a large variety of uses and applications in indexing, document annotation, searching, ranking, and other areas of document analysis and processing.

Brief Descri~tivn of the Drawing [Q021~ FIG.1 is block diagram of the software architecture of one embodiment of the pit invention.
[D022~ FIG. 2 illustrates a method of identifying phrases in documents.
[0023] FIG. 3 illustrates a document with a phrase window and a secondary windolw_ [0024) FIG. 4 illustrates a method of identifying related phrases.
[0025 FIG. 5 illustrates a method of indexing documents for related phrases.
[002b~ FIG. 6 illustrates a method of retrieving documents based on phrases.
[0027 FIG. 7 illustrates operations of the preser~tation system to present search result.
[002$) FIGS. 8a and 8b illustrate relationships between referenca~ng and referenced documents.
j0029~ The figures depict a preferred embodiment of the pxesent invention for purposes of illustration only. Gne skilled in the art will readily recognize from the followqng discussion that alteniative embodiments of the structures and methods Mlustr~ted herein may be employed without departing from the principles of the inven#on described herein.
Detailed Description of the Invention I. System Overview [0030 Referring now to FIG.1, there is shown the software architecture of an emboaiment of a search system 1U0 in accordance with one embodiment of present invention. In this embodiment, the system includes a indexing system 110, a search system 1120, a presentation system 130, and a front end server I40.
[003Ij The indexing system 110 is responsible for identifying phrases in documents, and indexing docux~nents according to their phrases, by accessing various websit,~s 190 and other document collec6ions. The front end server 14Q
receives queries from a user of a client 170, and provides those queries to the search system 120. The search system 120 is responsible for searching far documents relevant to the search query (search results), including identifying any phrases in the search query, and then ranking the documents in the search results using the presence of phrases to influxnce the ranjCing order. The search system 120 provides the search xesults to the presentation system 130. The presentation system 130 is responsible for modifying the search results includif~g removing near duplicate documents, and generating topical descriptions of documents, and providing the modified search results back to the front end server 140, which pmvides the results to the client 270. The system 100 fcuther includes an index 250 thak stares the indexing information pertaining to documents, and a phrase data store 160 that stores phrases, and related statistical information.
j0I?32] Jn the context of this application, '.documents" are understood bo be any type of ~nedza that can be indexed and retrieved by a search engine, including web documents, images, multimedia files, text documents, PDFs or other image formatted files, and so forth. A document may have one or more pages, partitions, segments or outer cbrnponents, as appropriate to its content and type. Equivalently a document may be referred to as a "page,' as commonly used to refer to documents on the Internet No limitatit~n as to the scope of the invention is implied by the use of the generic term "documents." The search system 100 operates over a large corpus of documents, such as the Ini~ernet and World Wide Web, but can likewise be used in more limited collections, such a~ for the document collections of a library or private enterprises. In either context it will be appreciated that the documents are typically distributed across many different computer systems and sites. Without Ions of generality then, the documents generatiy, regardless of format or location (e.g., which website or database) will be collectively referred to as a corpus or document collection Each document has an associated identifier that uniquely identifies the document; the identifier is preferably a URL, lput other types of identifiers (e.g., document numbers) may be used as well. In this disclosure, the use of URLs to identify documents is assumed.
II. Indexing~~stem [0033]In one embodiment, the indexing system 120 provides three primary functidnal operations:1) identification of phrases and related phrases, 2) indexing of documents with respect to phrases, and 3) generation and maintenance of a phrase-based taxonomy. Those of skill in the art will appreciate that the indexing system 110 will perform other functions as well in support of conventional indexing functions, and thus these other operations are not further described herein. The indexing system 110 operates on an index 15U and data repository I60 of phrase data. These data repositories are fuzither described below.
1. Phrase Identification (0034]- The phrase identification operation of the indexing systenn 120 identifies "good's and "bad" phrases in the document collection that are useful to indexing and searchpng documents. In one aspect, good phrases are phrases that tend to occur in more than certain percentage of documents in the document collection, and/or are indicated as having a distinguished appearance in such documents, such as delimited by markup tags or other morphological, format, or grammatical markers. Another aspect of g~oo~i phrases is that they are predictive of other good phrases, and ate not merely sequertces of words that appear in the lexicon. For example, the phrase "President of the Unired'tates" is a phrase that predicts other phrases such as "George $ush"
and °Bill Clinta~." FIowever, othex phrases are not predictive, such as "fell down the stairs" or "top a:( the morning,' out of the blue,° since idioms and colloqc~isms like these tend fio appeax with many other different and unrelated phrases. Thus, the phrase iden 't~f~,'cation phase determines which phrases are good phrases and which are bad (i.e., lacking in predictive power).
[Or135]I Referring to now FIG. 2, the phrase identification process has the foIlowtng functional stages:
[0036]I 200: Collect possible and good phrases, along with frequency and co-occurrence statistics of the phrases.
(0037j1 202: Classify possible phrases to either good or bad phrases 'based on freqme~ecy statistics.
[0038) 204: Prune good phrase Iist based on a predictive measure derived from the co-pccurrence statistics.
[0039] Each of these stages will now be described in further detail.
[0040]' The first stage 200 is a process by which the indexing system 110 crawls a set of c~ocumenfis in the document collection, making repeated partitions of the document collection ever time. One parkition is processed per pass. The number of documents crawled per pass can vary, and is preferably about 1,000,000 per partition. It is preferred that only previously uncrawled documents are processed iun each partition, untcq att documents have been processed, or some other termination criteria is met In practice, the crawling continues as new documents are being continually added to the document collection. The following steps are taken by the indexing system 1I0 for each document that is crawled.
[0041] Traverse the words of the dorurnent with a phrase window length of n, where ms a desired maximum phrase length. The length of the window will typically be at least 2, and preferably 4 or b terms (words). Preferably phrases include all words in the phrase window, including what would otherwise be characterized as stop words, such asj"a°, "the," and 8o forth. A phrase window may be terminated by an end of line, a paragraph return, a markup tag, or other indicia of a change in content or format.
[0042) FIG. 3 illustrates a portion of a document 300 during a traversal, showing the phrase window 302 starting at the word "shock" and extending 5 words to the right.
The firsr word in the window 302 is candidate phrase i, and the each of the sequences i+I, i+2; i+3, i+4, and i+5 is likewise a candidate phrase. Thus, in this example, the candidate phrases are: "stock", "stock dogs", "shock dogs for", "stock dogs for the', "stock crags for the Basque", and "stock dogs for the Basque shepherds".
j0043J In each phrase window 302, each candidate phrase is checked in turn ha determ~jne if it is already present in the good phrase list 208 or the possible phrase list 206. If the candidate phrase is not present in either the good phrase Iist 208 or the possibt~ phrase list 2fl6, then the candidate has already been determined to be "bad" and is skipped.
(0044] If the candidate phrase is in the good phrase list 208, as entry gi, then the index 1~U entry for phrase g~ is updated to include the document (e.g., its URL or other document identifier), to indicate that this candidate phrase g~ appears in the current document. An entry in the index 150 for a phrase g,; (or a term) is referred to as the posting xist of the phrase g;. The posting list includes a list of documents d (by their document identifiers, e.g. a document number, or alternatively a URL} In which the phrase lxcurs.
[Oi145In addition, the co-occurrence matrix n2 is updated, as further explained below. !1n the very first pass, the good and bad Iists will be empty, and thus, most phrases will tend to be added to the possible phrase list 206.
[00461 If the candidate phrase is not in the good phrase List 208 then it is added to the possible phrase list 20b, unless it is already present therein. Each entry p on the possible phrase list 206 has three associated counts:
[0047] P(p): Nurnber of documents on which the possible phrase appears;
[0048] S{p): Number of all instances of the possible phrase; and [0049) M(p): Number of interesting instances of the possible phrase. An instance of a possible phrase is "interesting" where the possible phrase is distinguished from neighboring content in the document by grammatical or format markers, for example by being zr~ boldface, or underline, or as anchor text in a hyperlink, or in quotation marks.
These (~nd other) distinguishing appearances are indicated by various HTML
markup language tags and grammatical markers. These statistics are maintained for a phrase when id is placed on the goal phrase list 208.
j0050] In addition the various lists, a co-occurrence matrix 2I2 (G) for the good phrases is maintained. The matrix G has a dimension of m x m, where m is the number of gooq phrases. Each entry G(j, k) in the matrix represents a pair of good phrases (g;, gk). Tnc~ co-occurrence matxix 212 logically (though not necessarily physically) maintains three separate counts For each pair (ga, gk) of good phrases with respect to a secondary wind~.i~% 344 that is centered at fine current word i, and extends +/- h words. In one embodiment, such as illustrated in FIG. 3, the secandary wixidow 304 is 30 words. The co-occurrence matrix 212 thus maintains:
[~51~ R{jk): Raw Co-occurrence count. The member of times that phrase g~
appears in a secondary window 304 with phrase g~;
[0052jj D(j,k): Disjunctive Interesting count The number of times that either phrase g; or phrase g~ appears as distinguished text in a secondary window;
and [0053] C(jk): Conjunctive Interesting count; the number of times that both gf and phrase gx appear as distinguished text in a secondary window. The use of the conjunctive interesting count is particularly beneficial to avoid the circumstance where a phrasq (e.g., a copyright notice) appears frequently in sidebars, footers, or headers, and thus rs not actually predictive of other text (0054 Referring to the example of FIG. 3, assume that the "stock dogs" is on the good Iphrase list 208. as well as the phrases "Australian Shepherd" and "Australian Shepa#d~Qub of America". Both of these latter phrases appear within. the secondary windotw 304 around the current phrase "stock dogs". However, the phrase "Australian Shepn~rd Qub of America" appears as anchor text for a hyperlink (indicated by the underline) to website. Thus the raw co-occurrence count for the pair {"stock dogs", "Ausoqalian Shepherd"} is incremented, and the raw occurrence count and the disJun~tive interesting count for {"stock dogs", "Australian Shepherd Club of America"}
are bath incremented because the latter appears as distinguished text.
[0055]I The process of traversing each document with both the sequence window 302 a the secondary window 304, is repeated for each document in the partition.
[(?056J~ Once the documents in the partition have been traversed, the next stage of the indexing operation is to update 202 the good phrase list 208 from the possible phrase ~t 205. A possible phrase p on the possible phrase list 206 is moved to the good phrase ust 208 if the frequency of appearance of the phrase and the number of documents that the phrase appears in indicates that it has sufficient usage as semanticllly meaningful phrase, [0057] In one embodiment, this is tested as follows. A possible phrase p is removeid from the possible phrase list 206 and placed on the good phrase list 208 if [4058] a) P(p~ > 10 and S(p) > 20 (the number of documents containing phrase p is more~th,an 10, and.the number of occurrences of phrase p is more then 20);
or [4059] b) M(p) > 5 (the number of interesting instances of phrase p is more than

5) .
[0060] These thresholds are scaled by the number of documents in the partition;
for example if 2,000,000 documents are crawled in a partition, then the thresholds are approxjm.ately doubled. ~f course, those of skill in the art will appreciate that the specifiq values of the thresholds, or the logic of besting them, can be varied as desired.
[~6I] If a phrase p does not qualify for the good phrase list 208, then it is checked for qualification for being a bad phrase. A phrase p is a bad phrase if:
[0062] a) number of documents captaining phrase, P(p) < 2; and [0063] b) number of interestyng instances of phrase, M(pj = 0.
[0064] 'These conditions indicate that the phrase is both infrequent, and not used as indidatxve of significant content and again these thresholds may be scaled per number of documents in the partition.
[0065] It should be noted that the good phrase list 208 will naturally include individual words as phrases, in addition to mufti-word phrases, as described above.

This is~because each the first word in the phrase window 302 is always a candidate phrasal, and the appropriate instance counts will be accumulated. Thus, the indexing system 110 can automatically index both individual words (i.e., phrases with a single word f iat~d multiple word phrases. The good phrase list 208 will also be considerably shorted than the theoretical maximum based on all possible combinations of m phrases.
In typical embodiment, the good phrase list 208 will include about b.5x105 phrases. A
list o~ vad phrases is not necessary to store, as the system need only keep track of possiu~e and good phrases.
[006b1~ By the final pass through the document collection, the list of possible phrases will be relatively short, due to the expected distribution of the use of phrases in a lar~,~ ..orpus. Thus, if say by the I0~ pass (e.g.,10,000,000 documents), a phrase appe~a for the very first time, it is very unlikely fio be a good phrase at that time. It may be new phrase just coming into usage, and thus during subseduent crawls becomes increa~ngly common. In that case, its respective counts will increases and may ultimaitPty satisfy the thresholds for being a good phrase.
j0067~~ The third stage of the indexing operation is to prune 204 the good phrase list ~ using a predictive measure derived from the co-occurrence matrix 232.
Without prunulg, the good phrase list 208 is likely to include many phrases that while legit...,ately appearing in the lexicon, themselves do not sufficiently predict the presence of other phrases, or themselves are subsequences of longer phrases. Removing these weak good phrases results in a very robust likely of good phxases. To identify good phra , a predictive measure is used which expresses the increased likelihood of one phrase appearing in a document given the presence of another phrase. This is done, in one embodiment, as follows:

[006$1 As noted above, the co-occurrence matrix 212 is an m x m matrix of storing' data associated with the good phrases. Each row j in the matrix represents a goad x~~rase g~ and each column k represented a good phrase gk. For each good phrase g;, an e~cpected value E(gi) is computed. The expected value E is the percentage of d~un~nts in the collection expected to contain g;. This is computed, for example, as the ratio ~x the number of documents containing g~ to the total number T of documents in the collection that have been crawled: P(j)/T.
[0069] ~ .As noted above, the number of documents containing g; is updated each time g; jappears in a document. The value for E{g,} can be updated each time the counts for g~ ale incremented, or during this third stage.
[0070 Next, for each other good phrase gk (e.g., the columns of the matrix), it is determEmed whether g; predicts gk. A predictive measure for g~ is determined as follows:
[0071 i) compute the expected value E{gx). The expected co-occurrence rate E(jk) o~ g; and gx, if they were unrelated phrases is then E{gJ)*B(g~);
[007 ii) c4mpute the actual co-occurrence rate A(jk} of g~ and gk. This is the raw cc~toccurrence count R(j, k} divided by T, the total number of documents;
[0073] iii) g~ is said tn predict gk where the actual co-occurrence rate A(jk) exceed the expected co-occurrence rate E(jk) by a threshold a~naunt.
[0079 In one embodiment, the predictive measure is information gain. Thus, a phrase;g; predicts another phrase gx when the information gain I of gx in the presence of g; excefds a threshold. In one embodiment, this is computed as follows:
[oa75I I(j,k) = A(j,k)/E()~k) jOt)7Cs ~ i And good phrase g; predicts good phrase gk where:
1$

[0077] I(jk} > Information Gain threshold.
[0078] In one ernbodimen~, the information gain threshold is 1.5, but is prefera~Iy between 1.1 and 1.7. Raising the threshold ever 2.0 serves to reduce the possirniity that two otherwise unrelated phrases co-occwr more than randomly predicted.
10079] As noted the computation of information gain is repeated for each column k of the matrix G with respect to a given row j. Once a row is complete, if the information gain for none of the good phrases gk exceeds the information gain threshold, then tn~ means that phrase g; does not predict any other good phrase. In that case, gJ is remov~a from the good phrase Iist 208, essentially becoming a bad phrase. Note that the cal~mn j for the phrase g; is not removed, as this phrase itself may be predicted by other gpod phrases.
[0080] This step is concluded when all rows of the co-occurrence matrix 212 have been evaluated.
[0081] The final step of this stage is to prune the good phrase list 208 to remove incomplete phrases. An incomplete phrase is a phrase that only predicts its phrase extensWns, and which starts at the left most side of the phrase (i.e., the beginning of the phrase. The aphrase extension" of phrase p is a super-sequence that begins with phrase p. For example, the phrase "President of" predicts "President of the United States", "Presidlent of Mexico', "President of AT&T", etc. AlI of these latter phrases are phrase extensibns of the phrase "President of since they begin with "President ofri and are super-sjequences thereof.
[0082] Accordingly, each phrase gj remaining on the good phrase list 2ia8 will predictisome number of other phrases, based an the information gain threshold previol~sly discussed. Now, for each phrase g; the indexing systean 110 performs a string ~natrh with each of the phrases gk that is predicts. The string ma5ch tests whether each u~edicted phrase gk is a phrase extension of the phrase g;. If all of the predicted phra~__ gx are phrase extensions of phrase g~, then phrase gtis incomplete, and is remov,~d from the good phrase list 208, and added to an incomplete phrase list 216.
Thus, ~ there is at least one phrase gk that is not an extension of g;, then gJ is complete, and maintained in the good phrase list 208. For example then, "President of the Unitear- is an incomplete phrase because the only other phrase that it predicts is "Presiaent of the United States' which is an extension of the phrase.
j0083 The incomplete phrase list n6 itself is very useful during actual searching. When a search query is received, it can be compared against the incomplete phase list 216. If the query (or a portion thereof) matches an entry in the list, then the search isysfiem 120 can lookup the most likely phrase extensions of the incompiehe phrase (the phrase extension having the highest information gain given the incomplete phrase), and suggest this phrase extension to the user, or automatically search on the phrase extension. For example, if the search query is "President of the United," the search systernl 120 can automatically suggest to the user "President of the United States° as the search query.
j0084~ After the Iast stage of the indexing process is completed, the good phrase Iist 20~ will contain a large number of good phrases that have been discovered in the corpus Each of these goad phrases will predict at least one other phrase that is not a phrase;extension of it. That is, each good phrase is used with sufficient frequency and independence to represent meaningful coxucepts or ideas expressed in the corpus. Unlike existing systems which use predetermined or hand selected phrases, the good phrase list reflects~phrases that actual are being used in the corpus. Fcwther, since the above process, ~f crawling and indexing is repeated periodically as new documents are added to the apcument collection, the indexing system 120 automatically detects new phrases as they (enter the lexicon.
2. Identification of Related Phrases and Qusters of Related Phrases [008}, Referring to FIG. 4, the related phrase identification process includes the folIowmg functional operations.
[0086] 400: Identify related phrases having a high information gain value.
[00$7] 402: Identify clusters of related phrases.
[OUSB] 404: Store cluster bit vector and cluster number.
[0089] Each of these operations is now did in detail.
[0090] First, recall that the co-occum~ence matrix 212 contains good phrases g~, each of !which predicts at least one ether good phrase gx with an information gain greater than the information gain threshold. To identify 400 related phrases then, for each pair of good vhrases (g], gk) the information gain is compared with a Related Phrase threshold, e.g., 300. That is, gT and gx are related phrases where:
[0011] I(g~, gk) > 100.
[D092) This high threshold is used to identify the co-occurrences of good phrases that ark well beyond the statistically expected rates. Statistically, it means that phrases g;
and g~ Co-occur 100 times more than the expected co-occarrence xate. For example, given tt~e phrase "Monica Lewinsky~' in a document, the phrase "Bill Clinton"
is a 100 times x~.ore likely to appear in the same document, then the phrase "Bill Qinton" is likely to appear on any randornly selected document. Another way of saying this is that the accuracy of the predication is 99.99996 because the occurrence rate is 100:1.

I0493 Accordingly, any entry (g;, gk) that is less the Related Phrase threshold is zexoea~,.ut, indicating that the phrases g;, gk are not xelated. Any remaining entries in the co-~~-currence matrix 212 now indicate all relafied phrases.
[0494 The columns g~ in each row g; of the co-occurrence matrix 212 are then sorted by the information gain values I(g;, gk), so that the related phrase g~
with the highest information gain is listed first. This sorting thus identifies for a given phrase g;, which other phrases are most li"lcely related in terms of information gain.
[0095] The next step is to determine 402 which related phrases together form a dustQrlof related phrases. A cluster is a set of related phrases in which each phrase has high iru,'formation gain with respect to at least one other phrase. In one embodiment, clusters are identified as follows.
[4096] In each row g; of the matrix, there will be one ar more othex phrases that are related to phrase gf. This set is related phrase set Ry where R={gx, gr, ...g",}.
[0497 For each related phrase m in Ri, the indexing system 110 determines if each ofl the other related phrases in R is also related to ga. Thus, if I(g~;
g!) is also nom-zero, then g;, gk, and g7 are part of a cluster. This cluster best is repeated for each pair Sgt ~)~~~
[4098) For example, assume the good phrase "Bill Qinton" is related to the phrase$ "President", "Monica Ixwinsky", because the information gain of each of these phrases with respect to "Bill Qinton" exceeds the Related Phrase threshold.
Further assume that th.e phrase "Monica Lewinsky" is related to the phrase "purse designef'.
These phrases then form the set,R. To determine the clusters, the indexing system 110 evaluates the information gain of each of these phrases to the others by determining their cdrresponding information gains. Thus, the indexing system 110 determines the information gain I("President", "Monica Lewinsky"), I("President", "purse designer), and so forth, for all pairs in R. In this example, "Bill Clinton, "
"President", and "Monica Lewinsky"" form a one cluster, "Bill Clinton, " and "President" form a second cluster, and "Monic~ 'Lewir~sky" and "purse designer" form a third cluster, and "Monica Lewinsky", '.Bill Clinton," and "purse designer" farm a fourth cluster. This is because while "Bill Clintor~~ does not predict "poise designer" with sufficient information gain, "Monica Lewinsky" does predict both of these phrases.
]0099] To record 404 the cluster information, each cluster is assigned a unique cluster number (cluster ID). This information is then recorded in conjunction with each good phrase g,;.
[OIO~j 1n one embodiment, the cluster number is determined by a cluster bit vectoa~ that also indicates the orthogonality relationships between the phrases. The cluster bit vector is a sequence of bits of length n, the number of good phrases in the good phrase list 2fl8. For a given good phrase ~, the bit positions correspond to the sorted belated phrases R of gi. A bit is set if the related phrase gx in R is in the same cluster ~s phrase ~. More generally, this means that the corresponding bit in the duster bit vectpr is set if there is information gain in either direction between g;
and gx.
10101] 'The cluster number then is the value of the bit string that results.
This implementation has the property that related phrases that have multiple or one-way inform~.tion gain appear in the same duster.
]0102] An example of the duster bit vectors are as follows, using the above phrase:
Monica purse Quster Bill Clinton President Lewinsky desi~er ID

BiII ' 1 1 1 D 14 ton p~ t 1 1 0 0 12 Monijca pure desi er a 0 ? 1 3 (0103] To summarize then, after this process there will be identified for each good phrase g;, a set of related phrases R, which are sorted in order of information gain I(g~, gx) from highest to lowest. In addition, for each good phrase g;, there will be a cluster fit vector, the value of which is a cluster number identifying the primary cluster of which the phrase g; is a member, and the orthogonality values (1 or fl for each bit position) indicating which of the related phrases in R are in common clusters with g;.
Thus in~the above example, "Bill Clinton", "President", and "Monica Lewinsk~' are in cluster 24 based on the values of the bits in the row for phrase "Bill Clinbon".
(4104] To store this information, two basic representations are available.
First, as indicted above, the information may be stored ~ the co-occurrence matrix 212, wherein:
j0105] entry G[row j, col. k] _ (I~j,k), clusterNumber, clusterBitVector) (OIOb] Alternatively, the matrix representatioxi can be avoided, and all informaition stored in the good phrase list 208, wherein each row therein represents a good phrase g;:
[0107) Phrase rows = list [phrase g~ (I~j,k), clusterNumber, clusfierBitVector)j.
[0108] This approach provides a useful organization for dusters. First, rather than a sitrictly-and often arbitrarily-defined hierarchy of topics and concepts, this approa~n recognizes that topics, as indicated by related phrases, form a complex graph of relatpanships, where some phrases are related to many other phrases, and some phrased nave a more limited scope, and where the relationships can be mutual (each phrase~predicts the other phrase) or one-directional (one phrase predicts the other, but not vice versa). The result is that clusters can be charactPxized "local" to each good phrasel and some clusters will then overlap by having one or more common related P~-[OI09] For a given good phrase g~ then the ordering of the related phrases by inform~tian gain provides a taxonomy for naming the clusters of the phrase:
the cluster name is the name of the related phrase in the cluster having the highest information gain.
[0110] The above process provides a very robust way of identifying significant phrases that appear in the document collection, and beneficially, the way these related phrase$ are used together in natural "clusters" in actual practice. As a result, this data-driven iclustering of related phrases avoids the biases that are inherent in any manually directed "editorial" selection of related teams and concepts, as is common in marry systems.
3. lndexine Documents with Phrases and Related Phrases --...
j01111 Given the good phrase list 208, including the information pertaining to reiated;phrases and clusters, the next functional operation of the indexing system 110 is to inde3c documents in the document collection with respect to the good phrases and cluster, and store the updated information in the index 150. FIG, 5 ~7lustrates this process, irn which there are the following functional stages for indexing a docunnent:
[0112] 500: Post document to the posting lists of good phrases found in the document.

[0113] 502: Update instance counts and related phrase bit vector for related phases!and secondary related phrases.
j0114) 504: Annotate documents with related phrase information.
j0115) 506: Reorder index entries according to posting Iist size.
j0116) These stages are now described in further~detail.
[0117] A set of documents is traversed or cxawled, as before; this may be the same ojr a different set of documents. For a given document d, traverse 500 the document word by word with a sequence window 302 of length n, from position i, in the manner described above.
]0118) In a given phrase window 3fl2, identify all good phrases in the window, staxtxng at position i. Each good phrase is denoted as g~. Thus, g1 is the first good phrase, g2 would be the second good phrase, and so forth.
]0119] For each good phrase gi (example g2 "President" and g4 "President of ATT") host the document identifier (e.g., the URL) to the posting list for the good phrase g; in the index 150. This update identifies that the good phrase gi appears in this specific documgnt.
[0120) In one embodiment, the posting list for a phrase g; takes the following logical dorm:
[0I21] Phrase gr: list: (document d, Ilist: related phase counts] [related phrase infOrm~tion]) [0122) For each phrase g; there is a list of the documents d on which the phrase appear. For each document, there is a list of counts of the number of occurrences of the relatedpphrases R of phrase g; that also appear in document d.

(0123] In one embodiment, the related phrase information is a related phase bit vector. This bit vector may 'be characterized as a "bi-bit" vector, in that for each related phrase .gk there are two bit positions, g~-1, gr~2. The first bit position stores a flag indicating whether the related phrase gx is present in the document d (i.e., the count for g,~ in document d is greater than 0). The second bit position stores a flag that indicates whether a related phrase gi of gk is also present in document d. The related phrases g~ of a relat~ci phrase gk. of a phrase g~ are herein called the "secondary related phrases of g; "
The colrxits and bit positions correspond to the canonical order of the phrases in R
(sorted in order of decreasing information gain). This sort order has the effect of making the related phrase gx that is most highly predicted by g- associated with the most significant bit of the related phrase bit vector, and the related phrase gi that is least prediciied by g~ associated with the least signifiicant bit.
(0124] It is useful to note that for a given phrase g, the length of the related phrase bit vecfior, and the association of the related phrases to the individual bits of the vector, vv01 be the same with respect to all documents containing g. This implementation has the property of allowing the system to readily compare the related phrase;bit vectors for any {or all) documents containing g, to see which documents have a giver[ related phrase. This is beneficial for facilitating the search process to identify docurnlants in response to a search query. Accordingly, a given document will appear in the posting lists of many different phrases, and in each such posting list, the related phrase:vector for that document will be specific to the phrase that awns the posting list.
This as~~ect preserves the lcxality of the related phrase bit vectors with respect to individual phrases and documents.

[0125] Accordingly, the next stage 502 includes traversing the secondary windov~ X04 of the current index position in the document (as before a secondary window of +/_ IC terms, for example, 3U terms), for example from i-K to i+K.
For each related phrase g~ of gi that appears in the secondary window 304; the indexing system 11Q inc~ments the count of gk with respect to document d in the related phrase count. 1f g~ appears later in the document, and the related phrase is found again within the later secondary window, again the count is incremented.
[0126] As noted, the corresponding first bit grrl in the related phrase bit map is set based on the coVnt; with the bit set to 1 if the count for gx is >0, or set to 0 if the count equals 4.
[0127] Next, the second bit, gx-2 is set by looking up relataed phrase gx in the index 1~0, identifying in gk s posting list the entry for document d, and then checking the secondary related phrase counts (or bits) for gx for any its related phrases.
If any of these secondary related phrases counts/bits are set, then this indicates that the secondary related phrases of g; are also present in document d.
[0128] When document d has been completely processed in this manner, the indexing system 110 will have identified the following:
[0129] l) each good phrase g; in document d;
[0I30] ii) for each good phrase g~ which of its related phrases gx are present in docum$nt d;
[0131] iii) for each related phrase g~ present in document d, which of its related phrases gr ithe secondary related phrases of g;) are also present in document d.

a) Determinu~ the Topics for a Document [0132j The indexing of documents by piwases and use of the clustering informa~ti4n provides yet another advantage of the indexing system 120, which is the ability to determine the topics that a document is about based on the related phrase information.
]0133] Assume that for a given good phxase gi and a given document d, the posting Iist entry is as follows:
[0134j g~: document d: related phrase counts = {3,4,3,0,0,2,1,1,0}
[07.35] related phrase bit vector:={ 111110 00 00101010 01}
(0136] where, the related phrase bit vector is shown in the bi-bit pairs.
[0137] From the related phrase bit vectaor, we can determine primary and secondary topics for the document d. A primary topic is indicated by a bit pair (1,1), and a secaniiary topic is indicated by a bit pair (1,0). A related phrase bit pair of (1,1) indicatES that both the related phrase gk for the bit pair is present in document d, along the secqndary related phrases gr as well. This may be interpreted to mean that the author of the document d used several related phrases g;, g~, and gi together in drafting the document. A bit pair of (1,0) indicates that both g~ and gk are present, but no further second$ry related phrases from gE are present, and thus this is a less signiCxcant topic.
b) Document Annotation for Improved Ranking [0138] A further aspect of the indexing system 110 is the ability to annotate each dqcument d during the indexing process with information that provides for improved ranking during subseduent searches. The annotation process 506 is as follows;

(~1~9] A given document d in the document collection may have some number of outyi#~ks to other documents. Earn outlink (a hyperlink) includes anchor text and the document identifier of the target document. For purposes of explanation, a current document d being processed will be referred to as URLO, and the target document of an outlink on document d will be referred to as URLl. For later use in ranking documents in seardh results, for every link in URLO, which paints to some other URLi, the indexing system 110 creates an outlinlC score for the anchor phrase of that link with respect to URLO, ~d an inlink score for that anchor phrase with respect to URLi. That is, each link in the document collection has a pair of scores, an outlink score and an inlink score.
These slcores -are computed as follows.
(0140] ~n a given document URLO, the iztdexing system 110 identifies each outlink' fio another document URL1, in which the anchor text A is a phrase in the good phrase fist 208. FIG. 8a illustrates schematically this relationship, in which anchor text "A" in document URLO is used in a hyperlink 800.
(d141] In the posting list for phrase A, URL4 is posted as an outlink of phrase A, and U>~,L1 is posted as an inlink of phrase A. For URLO, the related phrase bit vector is completed as described above, to identify the related phrases and secondary related phrases of A present in URLO. This related phrase bit vecfior is used as the outlink score for the fink from URLO to URL2 containing anchor phrase A.
(0I42] Next, the inlink scone is determined as follow. For each inlink to URL1 containing the anchor phrase A, the indexing system 110 scans URLI, and determines whether phxase A appears in the body of URLI. If phrase A not only points to URLl (via a qutlink on URLO), but also appears in the content of URLI itself, this suggests that URL1 aan be said to be intentionally related to the concept represented by phrase A.

FIG. 8b illustrates this case, where phrase A appears in both URLO (as anchor text) and in the body of URL1. In this case, the related phrase bit vector for phrase A
for URL1 is used as the inlink score~for the Iink from URLO to URL1 containing phrase A.
[0148] If the anchor phrase A does not appear in the body of URL1 (as in FIG.
8a), thexi a different step is taken to determine the inlink score. In this case, dte indexing system 110 creates a related phrase bit vector for URL1 for phrase A
(as if phrase ,d~ was present in URL1) and indicating which of the related phrases of phrase A
appear In URL2. This related phrase bit vector is then used as the inlink score for the link fro~pn URLO to URL1.
[0144] For example, assume the following phrases are initially present in URLO
and UR~,2:
A~hor Phrase Related Phrase Bit Vector Australian blue red agility J~ocum' She herd Aussiemerle merle tricolortrainin t [UI45] (In the above, and following tables, the secondary related phrase bits are not shoal n). Tie URLO row is the outlink score of the link from anchor text A, and the URLi raw is the inlink score of the link. Here, URLO contains the anchor phrase "Austr~liart Shepard" which targets URL1. Of the five related phrases of "Australian Shepard". only one, "Aussie" appears in URLO. Intuitively then, URLO is only weakly about Aiustralian Shepherds. URLT, by comparison, not only has the phrase "Australian Shephetd" present in the b~ly of the document, but also has many of fine related phrased present as well, "blue merle," "xed merle," and "tricolor."
Accordingly, because the anchor phrase "Australian Shepard" appears in both URLO and URLI, the outlink;score for URLO, and the inlink score for URL1 are the respective rows shown above.
(0346] The second case described above is where anchor phrase A does not appear an URL1. in that, the indexing system 110 scans URL1 and determines which of the Telaf ed p~lraS4'& "A~1$$Ie, " "Mlle I11P~'le," "red merle," "tllCOlor; , and "agliity training' are present in URL1, and creates an relahed phrase bit vector accordingly, for example;
Anchor Phrase Related Phrase Bit Vector Australian blue red agility Document She herd Aussiemerle merle tricolortrainiil URL1 ~ 0 0 1 1 1 0 ]014T] Here, this shows that the URL2 does not contain the anchor phrase "Austr~Iian Shepard", but does contain the related phrases "blue merle", °red merle", and "tricolor".
10148; This approach has the benefit of entirely preventing certain types of manipcllations of web pages (a class of documents) in order to skew the results of a search. Search engines that use a ranking algorithm that relies on the number of links that point to a given document in order to rank that document can be. "bombed"
by artificially creating a large number of pages with a given anchor text which then point to a desired page. As a result, when a search query using the anchor text is entered, the desired page is typically returned, even if in fact this page has little or nothing to do with the anchor text Importing the related bit vector from a target document URL1 into the phriase A related phrase bit vector for document URLO eli>ninates the reliance of the search system on fast the relationship of phrase A in URLO painting to LTRL1 as an indicat~Or of significance or LTRL2 to the anchor text phrase.
(019] Each phrase in the index 150 is alsa given a phrase number, based on its frequency of occurrence in the corpus. The more common the phrase, the lower phrase number it receivesorder in the index The indexing system 110 then sorts 506 all of the posting lists in the index 150 in declining order according to the number of documents lisbedp~hrase number of in each posting list, so that the most frequently occurring phrased are listed first. The phrase number can then be used to look up a particular phrasey III. Search S3rstem [0150] The search system 120 operates to receive a query and search for documents relevant t~o the query, and provide a list of these docimnents (with links to the documents) in a set of search results. FIG. 6 illustrates the main functional operations of the search system 120:
(0151]' 600: Identify phrases in the query.
j0152a 602: Retrieve documents relevant to query phrases.
j0153] 604: Rank documents in search results according to phrases.
[0154] The details of each of these of these stages is as follows.
1. Identification of Phrases in the Query and Quer~r Expansion [0155] The first stage 600 of the search system 120 is to identify any phrases that are present in the query in order to effectively search the index. The following terminptogy is used in this section:
(0156] q: a query as input and receive by the search system 220.
[0157] Qp: phrases present in the query.

(0158] Qr: related phrases of Qp.
(0159] Qe: phrase extensions of Qp.
[016U1 Q: the union of Qp and Qr.
j0161j A query cI is received from a client 190, having up to some maximum numbelr of characters oz words.
[0162] A phrase window of size N (e.g., 5) is used by the search sys#ena 120 to traverse the terms of the query q. The phrase window starts with the first term of the query, extends N terns to the right. This window is then shifted right M N
times, where ~vI is the number of terms in.the query.
j0163] At each window position, there will be N terms (or fewer) terms in the windoyv. These terms constitute a possible query phrase. The possible phrase is looked up in tie good phrase Iist 208 to determine if it is a good phrase or not. If the possible phrase; is present in the good phrase Iist 208, then a phrase number is returned fox phraseY the possible phrase is now a candidate phrase.
[0164] After all possible phrases in each window have been tested to determine if fihey;are good candidate phrases, the search system 120 will have a set of phrase rtumb~rs for the corresponding phrases in the query. These phrase numbers are then sorted ,(declining order).
j0165). Starting with the highest phrase number as the first candidate phrase, the searchsystenn.120 determines if there is another candidate phrase within a fixed nume~cal distance within the sorted list, i.e., the difference between the phrase reumbers is withjin a threshold amount, e.g. 20,000. If so, then the phrase that.is leftmost in the query ~.s selected as a valid query phrase Qp. This query phrase and all of its sub-phrase$ is rennoved from the list of candidates, and the list is resorted and the process repeated. The result of this.process is a set of valid query phrases Qp.
j016~] For example, assume the search query is "Hillary Rodham Clinton Bill on the Senraae Floor". The search system 120 would identify the following candidate phrase, "Hillary Rodham Qinton Bill on," "1-iivaty Rodham Clinton Bill,' and "Hillary Rodha~ Qinton". The first two are discarded, and the Iast one is kept as a valid query phrase ~ Next the search system 124 would identify "Bill on the Senate Floor", and the subsptu9rases "Bill on the Senate", "Bill on the", .'Bill on", "Bill", and would select "Bill"
as a valid query phrase Qp. Finally, the search system 220 would parse "on the senate floor!' end identify "Senate Floor" as a valid query phrase.
[B'f 67j Next, the search system 120 adjusts the valid phrases Qp for capitalization. When parsing the query' the search system 124 identifies potential capita~izatiuns in each valid phrase. This may be done using a table of known capitailzations, such as "united states" being capitalized as "United States", or by using a grammar based capitalization algorithm. This produces a set of properly capitalized query phrases.
[018] 'The search system 120 then makes a second pass through the capitalized phrase, and selects only those phrases are leftmost and capitalized where both a phrase and its asubphrase is present in the set. For example, a search an "president of the united states" will be capitalized as "President of the United States'.
[01&9] Tn the next stage, the search system 120 identifies b42 the documents that are relevant to the query phxases Q. The search system 124 then retrieves the posting lists ~~.he query phrases Q and intersects these lisfis to determine which documents appears on the all (or some member) of the posting lists for the query phrases. if a phrase Q in the query has a set of phrase extensions Qe {as further explained below), then the search system 12Q first #orms the union of the posting lists of the phrase extensions, prior to doing the intersection with the posting lists. The search system 120 identifies phrase extensions by looking up each query phrase Q in the incomplete phrase Iist 21d, as descrined above.
[0170] 'The result of the intersection is a set of documents that are relevant to the query. indexing documents by phrases and related phrases, identifying phrases Q in the query, end them expanding the query to include phrase extensions results in the selecteon of a set of documents that are more relevant to the query then would result ixa a conventi.xnal Boolean based search system in which only documents that contain the query terms are selected [0171] In one embadiment, the search system 220 can use an optimized mechanism to identify documents responsive to the query without having to intersect all of the ppsting lists of the query phrases Q. As a result of the structure of the index 150, for eacr~ phrase g;, the related phrases gx are known and identified in the related phrase bit vectt~r for gx. Accordingly, this information can be used to shortcut the intersection process; where two or more query phrases are related phrases to each other, or have common related phrases. rn those cases, the related phrase bit vectors can be directly accessed, and then used next to retrieve corresponding documents. This process is more fully described as follows.
[0172] Given any two query phrases Q1 and QZ, there are three possible cases of relations:
[0173] 1) Q2 is a related phrase of Q1;

[074) 2} Q2 is not a related phrase of Q1 and their respective related phrases Qr1 arid Qr2 do not intersect {z.e., no common related phrases); and [015] 3} Q2 is not a relafied phrase of Ql, but their respective related phrases Qr1 aid Qr2 do intersect.
I017b], For each pair of query phrases the search system ?20 determines the appropriate case by looking up the related phrase bit vector of the query phxases Qp.
j01T7]I The search system ?2(1 proceeds by retrieving the posting list for query phrasal Q1, which contains the documents containing Q?, and for each of these doc~menfs, a related phrase bit vector. The related phrase bit vector for Ql will indicated whether phrase Q2 {and each of the remaining query phrases, if any}
is a related phrase of Q1 and is present in the document.
[0178] If the first case applies to Q2, the search system 120 scans the related phrase nit vector for each document d in Q?'s posting list to determine if it has a bit set for Q2~ 1f this bit is nvt set in for document d in Q1's posting list, then it means that Q2 does npt appear in that document. As result, this document can be immedia#ely eliminated from further consideration The remaining documents can then be scored.
'This means further that it is unnecessary for the search system 120 to process the posting lists of~Q2 to see which documents it is present in as well, thereby saving compute time.
I0179j~ if the second case applies to Q2, then the two phrases are unrelated to each other. For example the query "cheap bolt action rifle" has two phrases 4cheap" and "bolt alction rifle". Neither of these phrases is related to each other, and further the related phrases of each of these do not overlap; i.e., "cheap" has related phrases "low cost," 'inexpensive," "discount," "bargain basement,' and "lousy,", whereas "bolt action rifle" has related phrases "gun, " "22 caliber", "magazine fed,' and "Armalite AR

30M", v~hich Iists thus do not intersect. In this case, the search system 320 does the regular intersection of the posting lists of Q1 and Q2 to obtain the documents for scoring.
[0180j If the third case applies, then here the two phrases Q1 and Q2 that are not related, ;but that do have at least one related phrase in common. For example the phrases "bolt action rifle" and "2Z° would both have "gun" as a relafied phase.
In this case, the search slystem 120 retrieves the posting lisps of both phrases Ql and Q2 and intersects the Iiststo produce a list of documents that contain both phrases.
[~181j The search system 120 can then quickly scare each of the resulting docwne~nts. First, the search system 120 determines a score adjustment value for each docunnqnt. The score adjustment value is a mask formed from the bits in the positions corresponding to the query phrases Q1 and Q2 in the related phrase bit vector for a dorumsnt. For example, assume that Q1 and Q2 correspond to the 3~ and 6~ bi-bit positiorjs in the related phrase bit vector for document d, and the 'bit values in 3rd position are (1,1) and the bit values in the 6~ pair are (1,0), then the score adjustment value is~the bit mask "00 0011 00 0010". The score adjustment value is then used to mask th!e related phrase bit vectar for the documents, and modified phrase bit vectors then ark passed into the ranking function (next described) to be used in calculating a body scbre for the documents.
2. Ranting a) Ranking Documents Based on Contained Phrases j0182] The search system 320 provides a ranking stage 604 in which the documents in the search results are ranked, using the phrase information in each document's related phrase bit vector, and the cluster bit vector far the query phrases.

'This approach ranks documents accozding to the phrases that axe contained in the documlent, or informally "body hits."
[0183j As descn'bed above, for any given phrase g~, earh document d in the ~'s posting list has an associated related phrase bit vector that identifies which relahed phrased gr and which secondary related phrases g~ are present in document d.
The more related vhrases and secondary related phrases present in a given document, the more bits that wiI( be set in the document's related phrase bit vector for the given phrase. The more its that are set, the greater the numerical value of the related phrase bit vector.
[0184] Accordingly, in one embodiment, the search system 120 sorts the documlents in the search results accordnag to the value of their related phrase bit vectors.
The documents containing the most related phrases to the query phrases Q will have the highest valued related phrase bit vectors, and these documents will be the highest ranking documents in the search results.
[0185] ~ This approach is desirable because semantically, these documents are most i~pically relevant to the query phrases. Note that this approach provides highly releva~et documents even if the documents dv not contain a high frequency of the input query terms q, sixtce related phrase information was used to both identify relevant documlents, and then rank these documents. 'Documents whiz a low frequency of the input query terms maystill have a large number of related phrases to the query terms and phrases and thus be more relevant than documents that have a high frequency of just th~ query terns and phrases but no related phrases.
j018Gj In a second embodiment, the search system 120 scores each document in the resort set according which related phrases of the query phrase Q it contains. This is done a~ follows:

[01871" Given each query phrase Q there will be some number N of related phrases Qr to the query phrase, as identified during the phrase identification process.
As described above, the related query phrases Qr are ordered according to their inforn~atian gain from the query phrase Q. These related phrases are then assigned f pointsa started with N pointy for the first related phrase Qr2 (i.e., the related phrase Qr with tt~e highest information gain from Q), then N-1 points for the next related phrase Qr2, then N-2 points for Qr3, and so o~ sa that the last related phrase QrN is assigned 1 point.
[078$y Each document in the search results is then scored by determining which related phrases Qr of the query phrase Q are present, and giving the document the points assigned to each such related phrase Qr. The documents are then sorted from highest to lowest score.
[OI89j. As a further refinement, the search system 120 can cull certain documents from tie result set. 7n some cases documents may be about many different topics; this is particittarly the case for longer documents, In many cases, users prefer documents that are strspngly on point with respect to a single topic expressed in the query over docun~ients that are relevant to many different topics.
[0190 To cull these latter types of documents, the search system 120 uses the clusteynformatian in the cluster bit vectors of the query phrases, and removes any documEent in which there are more than a threshold number of clusters in the document.
For example, the search system 120 can remove any documents that contain more than two cl~csters. This clustier threshold can be predetermined, or set by the user as a search parameter.

b) RardCing Documents Based on Anchor Phrases [019I] In addition to ranking the documents in the search results based on body hits of cjuery phrases Q, in one embodiment, the search system I20 also ranks the documents based on the appearance of query phrases Q and related query phrases Qr in anchors m other documents. In one embodiment, the search system 120 calculates a score for each document that is a tuncfion {e.g., linear combination) of two scores, a body hid score and an anchor hit score.
[0192] For example, the document score for a given document can be calculated as folic=:
[0193] Score = .30x{body hit score)*.7Q"t{anchor hit score).
[0194] The weights of .30 and .70 can be adjusted as desired. The body hit score for a document is the numerical value of the highest valued related phrase bit vector for the document, given the query phrases Qp, in t3~e manner described above.
Alternatively, this value can directly obtained by the search system 120 by looking up each quiery phrase Q in the index 150, accessing the document from the posting list of the query phrase Q, and then accessing the related phrase bit vector.
[0195] The anchor hit score of a document d a function of the related phrase bit vectors bt the query phrases Q, where Q is an anchor term in a document that references document d. When the indexing system 110 indexes the documents in the document collection, it maintains for each phrase a list of the documents in which the phrase is anchor text in an outIink, and also for each document a list of the ir~links {and the associated anchor text) from other documents. The inlinks far a document are references {e.g. hy~erlinks) from other documents (referencing documents) to a given document.

[0196) To determine the anchor hit score for a given document d then, the search system $20 iterates over the set of referencing documents R (i~I to number of referenang documents) listed in index by their anchor phrases Q, and sums the followu~l; product:
[0197] R;.Q.Related phrase bit vector*D.Q.Related phrase bit vector.
[0198] The product value here is a score of how tropical anchor phrase Q is do document D. This score is here called the "inbound scare component." This product effectively weights the current document D's related bit vector by the related bit vectors of anchor phrases in the referencing document R. If the referencing documents R
themselves are related to the guery phrase Q (and thus, have a higher valued related phrase grit vector), then this increases the significance of the current document D score.
The boc~.y hit score and the anchor hit score are then combined to create the docu~aaent score, a~ described above.
[0199] Next, for each of the referencing documents R, the related phrase bit vector for each anchor phrase Q is obtained. This is a measure of haw topical the anchor phrase ~ is to the document R. This value is here called the outbound score component.
]0200] from the index 15Q then, all of the (referencing document, referenced document) pairs are extracted for the anchor phrases Q. These pairs. are then sorted by their associated (outbound score component, inbound score component) values.
Depentung on the implementation, either of these componenfis can be the primary sort key, ana the other can be the secondary sort key. The sorbed results are then presented to the uøer. Sorting the documents on the outbound score component makes documents that hake many related phrases to the query as anchor hits, rank most highly, thus represe#~ting these documents as "expert" documents. Sorting on fhe inbound document score makes documents that frequently referenced by the anchor terms the most High ranked, 3. Phrase $ased Persanalization of Search [0201] Another aspect of the search system 220 is the capability to personalize 6U6 orcustomize the ranking of the search results in accordance with a model of the user's particular interests. In this manner, documents that more likely to be relevant to the user's interests are ranked higher in the search results. Tile personalization of search resort ais as follows.
[0202] As a preliminary matter, it is useful to define a user's interests (e.g., a user mode>~ in terms of queries and documents, both of which can be represented by phrases.
For ark input search query, a query is represented by the query phrases Q, the related phrases of Qr, and phrase extensions Qe of the query phrases Qp. This set of terms and phrases thus represents the meaning of the query. Next, the meaning of a document is represented by the phrases associated with the page. As described above, given a query and document, the relevant phrases for the docwment are determined from the body scores(the related bit vectors) for all phrases indexed to the document.
Finally, a user can be! represented as the union of a set of queries with a set of documents, in terms of the pturases that represent each of these elements. The particular documents to include in the set representing the user can be determixted from which documents the user selects in previous search results, or in general browsing of the corpus (e.g., accessing documents on the Internet), using a client side tool which monitors user actions and destinations.
[0203 The process of constructing and using the user model for personalized rankir(g is as follows.

[02041 First, far a given user, a list of the last K queries and P documents accessed is maintained, where K and P are preferably about 250 each. The lists may be mainta~ed in a user account database, where a user is recognized by a login or by browsed vookies. For a given user, the lists wilt be empty the first time the user provides a query;
[0205] Next, a query q is received from the user. The related phrases Qr of q are retrieve, along with the phrase extensions, in the manner described above.
This forms the query model.
[0~?6] In a first pass (e.g., if there are no stored query information for the user}, the search system 120 operates to simply return the relevant documents in the search result t0 the user's query, without further customized rankixig.
[0207] A client side browser tool monitors which of the documents in the search results the user accesses, e.g., by clicking on the document link in the search results.
These accessed documents for the basis for selecting which phrases will become part of the used model. For each such accessed document, the seaxch system 120 retrieves the document model for the document, which is a list of phrases related to the document.
Each phrase that is related to the accessed document is added to the user model.
[0208] Next, given the phrases related to an accessed document, the clusters associa~ea with these phrases can be determined from the cluster bit vectors far each phrase. 'For each cluster, each phrase that is a member of the cluster is determined by looking the phrase up in its related phrase table that contains the cluster number, or cluster ~rt vector representation as described above. This cluster number is then added to the user model. In addition, for each such cluster, a counter is maintained and incremented each time a phrase in that duster is added to the user model These counts may b~ used as weights, as described below. Thus, the user model is built from phrases included in clusters that are present on a document that the user has expressed an interest in by accessing the document [02~9j The same general approach can be more precisely focused to capture phrase information where a higher level of ini~rest than merely accessing the document is atarnfested by the user (which the user may do simply to Judge if indeed the document is relevant). For example, the collection of phrases into the user model may be limited ho those documents that the user has printed, saved, stored as a favorite or link, e~nail to another user, or maintained open in a browser window for an extended period of time (e.g.,10 minutes). These and other actions manifest a higher level of interest in the document.
[0210j When another query is received from the user, the related query phrases Qr areretrieved. These related query phrases Qr are intersected with the phrases listed in the user model to determine which phrases are present in both the query and the user modell A mask bit vector is initialized for the related phrases of the query Qr. This bit vector its a bi-bit vector as descaribed above. For each related phrase (fir of the query that is also present in the user model, both of the bits for this related phrase are set in the mask tit vector. The mask bit vector thus represents the related phrases present in both the query and the user model.
[0211j~ The mask bit vector is then used to mask the related phrase bit vector for each dlocument in the current set of search iesults by ANDing the related phrase bit vector;with the mask bit vector. This has the effect of adjusting the body score and the anchoX nit score by the mask bit vector. The documents are then scored for their body score end anchor score as before and presented to the user. This approach essentially required that a document have the query phrases that are included in the user model in order tci be highly ranked.
[0212j As an attemative eu~bodiment, which does not impose the foregoing tight constant, the mask bit vector can be cast into array, so that each bit is used to weight the cluster counts for the related phrases in the user model. Thus, each of the cluster counts bets multiplied by 0 or 1, effectively zeroing or maintaining the counts. Next, these counts themselves are used as weights are also used to multiply the related phrases for each document that is being scored. This approach has the benefit of allowinig documents that do not have the query phrases as related phrases to still score appropriately.
[0213] Finally, the user model may be limited do a current session, where a sessionas an interval of time for active period of time in search, after which session the user mcSdel is dumped. Alternatively, the user model for a given user may be persisted over ti~~t~p, and then down-weighted or aged.
iV. Result Pregentation I0214j The presentation system 130 receives the scored and sorted search results from 1 search system 120, and performs further organizational, annotation, and clusterilng operations prior to presenting the results to the user. These operations facilitate the user s understanding of the content of the search results, eliminate duplicaites, and provide a more representative sampling of the search results.
FIG. 7 illustrates the main functional operations of the presentation system 130:
[0215] 700: Quster documents accord'mg to topic clusters [0216j 742: Generate document descriptions [0217. 704: Eliminate duplicate documents.
[02T8] Each of these operations takes as an input the search results 701 and outpuijs modified search results 703. As suggested by FIG. 7, the order of these operar~ons is independent, and may be varied as desired for a given embodiment and thus tile inputs may be pipelined instead of being in parallel as shown.
1. Dynamic Taxonomy Generation for Presentation [0219]! For a given query, it is typical to return hundreds, perhaps even thousalnds of documents that satisfy the query. In many cases, certain documents, while having different content from each other, are sufEciently related to form a meaningful group of related documents, essentially a cluster. Most users however, do not review beyond the first 30 or 40 documents in the search results. Thus, if the first docu ants for example, would come from three clusters, but the next 100 documents repr2,~~tt an additional four clusters, then without further adjustment, the user will typicahy riot review these later documents, which in fact may be quite relevant to the user's query since they represent a variety of different topics related to the query. Thus, it is here desirable to provide the user with a sample of documents from each cluster, therebty exposing the user to a broader selection of different documents from the search results. The presentation system X30 daes this as follows.
[oz2al~ As in other aspects of the system 100, the presentation system 130 makes use of the related phrase bit vector for each document d in the search results. More specifically, for each query phrase Q, and for each document d 3n Q's posting list, the related phrase bit vector indicates which related phrases Qr are present in the document.
Uver the set of documents in the search results then, for each related phrase Qr, a count is determined for how many documents contain the related phrase Qr by adding up the bit vai~es in the bit position corresponding to Qr. When summed and sorted over the searcn;results, the most frequently occurring related phrases Qr will be indicated, each of wh will be a cluster of documents. The most frequently occurring related phrase is the first cluster, which takes as its name its related phrase Qr, and so on for the top three to five'clusters. Thus, each of the top clusters has been identified, along with the phrase Qr as a name or heading for the cluster.
[02211 Now, documents from each cluster can be presented to the user in variou,'s ways. In one application, a fixed number of documents from each cluster can be presented, for example, the 10 top scoring documents in each duster. In another application, a proportional number of documents from each cluster may be presented.
Thus, tt there are 100 documents in the search result, with 50 in cluster 2, 30 in cluster 2, in cluster 3, 7 in cluster 4, and 3 in cluster ~, and its desired to present only 20 documents, then the documents would be select as follows: 10 documents from cluster 1; y documents from cluster 2, 2 documents from cluster 3, and 1 document from chrster 4. Thd documents can then be shown to the user, grouped accordingly under the appropriate cluster names as headings.
[0222] For example, assume a search query of "blue merle agility Wig", for which 'the search system 120 retrieves 100 documents. The search system 120 will have already identified "blue merle" and "agility training" as query phrases. The related phrases of these query phrases as:
[0223) "blue merle":: "Australian Shepherd," °red merle," Ntricolor,"
"aussie°;
[0224]~ "agility training":: "weave poles," "teeter," "tunnel," "obstacle,"
"border collie"!

[0225T The presentation system 130 then determines for each of the above related~phrases of each query phrase, a count of the number of documents contain such phrase. roc example, assume that the phrase weave poles" appears in 75 of the documents, "teeter'' appears in 60 documents, "red merle" appears in 50 documents.
Then ttie first cluster is named "weave polesN and a selected number of documents from that clulster are presented; the second cluster is named "teeter," and sel~ted number are presented as well, and so forth. For a fixed presentation,10 documents from each cluster tray be selected. A proportional presentation would use a proportionate number of documents from each cluster, relative to the hotal number of documents.
2. Topic Based Document Descriptions j022fi] A second function of the presentation system 130 is the creation 702 of a document description that can inserted into the search result presentation for each document These descriptions are based on the related phrases that are present in each document, and thus help the user understand what the document is about in a way that is conte~ctualIy related to the search. The document descriptions can be either general or person~Iized to the user.
a} General Topic Document Descriptions [0227) As before, give a query, the search system 120 has determined the relahed iquery phrases Qr and the phrase extensions of the query phrases as weld, and then identified the relevant documents for the query. The presentation system accesses each document in the search resul#s and perform the follow operations.
[0228] First, the presentation system 130 xanks the sentences of the document by the nu~tber of instances of query phrases Q, related query phrases Qr, and phrase exfiensi!ans Qp, thereby maintaining for each sentence of a document counts of these three a$pects.
j0229) Then the senfiences are sorted by these counts, with the f~.rst sort key being the count of query phrases Q, the second sort key being the count of xelated query phrases fir, and the final sort key being the count of phrase ext~sions Qp.
j02ao~ ' Finally, the top N (e.g., 5~ sentences following the sort are used as the description of the document. This set of sentences can be formatted and inducted in the presentation of the document in the rnodified search results 703. 'This process is repeated for some number of documents in the search results, and may be done on demand each time the user requests a next page of the results.
b) Personalized Topic Based Document Descriptions [023T) In embodiments where personalization of the search results is provided, the doCUment descriptions can likewise be personalized to reflect the user interests as expressled in the user model. The presentation system 130 does this as follows.
j0232) First, the presentation system 230 determines, as before, the related phrase8 that are relevant to the user by intersecting the query related phrases Qr with the use#' model (which lists the phrases occurring in documents accessed by the user).
[0233) The presentation system 130 then stable sorts this set of user related phrased Ur according to the value of the bit vectors themselves, prepending the sorted list to tote list of query related phrases (fir, and removes any duplicate phrases. The stable sort maintains the existing order of equally ranked phrases. This results in a set of related ~rhrases which related to the query or the user, called set Qu.
[D234) Now, the presentation system 230 uses this ordered list of phrases as the basis fot' ranking the sentences in each document in the search results, in a manner si~la~ ~ the general document description process described above. Thus, for a given docwrcent, the presentation system 230 ranks the sentences of the document by the number of instances of each of the user related phrases and the query related phrases Qu, anjd sorts the ranked sentences according to the query counts, and finally sorts based nn the number of phrase extensions for each such phrase. Whereas previously the sort keiys where in the order of the query phrases Q, related query phrases Qr, and phrase extension Qp, here the sort keys are in the order of the highest to lowest ranked user re'Iated phrases Ur.
[0235] Again, this process is repeated for the dacuments in the search results (either: on demand or aforehand). For each such document then the resulting document description comprises the N top ranked sentences from the document. Here, these sentences will the ones that have the highest numbers of user related phrases Ur, and thus represent the key sentences of the document that express the concepts and topics most relevant to the user tat least according to the information captured in the user model).
3. Duplicate Document Detection and filimination [0236] In large corpuses such as the Internet, it is quite common for there to be multipXe instances of the same document, or portions of a document in many different locations. For example, a given news article produced by a news bureau such as the Associated Press, may be replicated in a dozen or more websites of individual newspa;~ers. Including all of these duplicate documents in response to a search query only burdens the user with redundant information, and does not usefully respond to the query. Thus, the presentation system 130 provides a further capability 704 to identify docuavents that are likely to be. duplicates or near duplicates of each other, and only incLud~ one of these in the search results. Consequently, the user receives a much more diversified and robust set of results, and does not have to waste time reviewing docxxm~ents that are duplicates of each other. The presentation system 130 provides the functidnaIity as follows.
[0237] The presentation system x34 pzocesses each document in the search result set 741; For each document d, the presentation system 130 first determines the list of relatet~ phrases R associated with the document. For each of these related phrases, the presentation system 130 ranks the sentences of the document according to the frequency of occurrence of each of these phrases, and then selects the top N (e.g., 5 to T4) ranlang sentences. This set of sentences is then stared in association with the document. C?ne way td do this is to concatenate the selected sentences, and then take use a hash table to store tlhe document identifier.
[02381 Then, the presentation system 130 compares the selected sentences of each document d to the selected sentences of the other documents in the search results 701, and if the seiected sentences match (within a tolerance), the documents are presumed to I~ duplicates, and one of them is removed from the search results.
For exaznp~e, the presentation system 130 can hash the concatenated sentences, and if the hash t~.ble already has an entry for the hash value, then this indicates that the current documient and presently hashed document 'are dupiicates. The presentation system 130 can then update the table with the document ID of one of the documents.
Preferably, the presentation system 130 keeps the document that has a higher page rank or other query independent measure of document sigttihcance. In addition, the presentation system 134 can modify the index 150 to remove the duplicate document, so that it wr~l not appear in future search results for any query.

(0239] The same duplicate elimination process may be applied by the indexing system 10 directl~.T. When a document is crawled, the above described document descriptionprocess is performed to obta~ the selected sentences, and then the hash of these shntences. If the hash table is filled, then again the newly crawled document is deemet~ to be a duplicate of a previous document. Again, the indexing system 1I0 can then keep the document with the higher page rank or other query independent measure.
(~24tt) The present invention has been aescxibed 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 programming 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 syskem may be implemented via a combination of hardware and software, as described, or entirely in hardware elements. Also, the particular division of fu~tionality between the various system components described herein is merely exemplary, and not manda.mry; functions performed bt~ a single system component may instead be performed by multiple components, and functiar~ performed bar multiple components may instead performed by a single component.
[0241) Some portions of above description present the features of die present inventikm 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 descn'bed functionally or logically, are understood to be implemented by computer programs.
Furtheratore, it has d.
J

also proven convenient at times, to refer to these arrangements of operations as modules or by few-=~ctional names, without loss of generality.
[t7242I Unless specifically stated otherwise as apparent from the above discus$ion, it is appreciated that throughout the description, discussions utilizing terms such a~ °processing" or ~~cOmputing" Or "calculating" Or '~detertnin~mg" Or "displaying"
or the like, refer to the action and processes of a computer system, or similar electronic computing device, that manipulates and transforms data represented as physical ~electrqnic) quantities within the computer system memories or registers or other such informtation storage, transmission or display devices.
[0245] Certain aspects of the present invention include process steps and instructions described herein in the form of an algorithm. It should be rioted that the process steps and instructions of the present invention could be embodied in software, firnaw~re or hardware, and when embodied in software, could be downloaded to reside on ands be operated from different platforms used by real time network operating system.
j0244~ The present invention also relates to an apparatus for performing the operat'~ns herein. This apparatus may be specially constructed for the required purposl~s, or it may comprise a general-purpose computer selectively activated or reconfig<u~ed by a computer program stored an a computer readable medium that can be accesseld 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 jdisks, optical disks, CD-ROMs, magnetic-optical disks, read-only memories (ROMs~, random access memories (RAMS), EPROMs, EEPROMs, magnetic or optical cards, application specific integrated circuits (ASICs), or any type of media suitable for storing electronic instructions, and each coupled to a computer system bus.
Furthe~nnore, the computers referred to in the specification may include a single processor or may be architectures employing multiple processor designs for increased computing capability.
[0245] The algorithms and operations presented herein are not inherently related~to any particular computer or other apparatus. Various general purpose systems may al~~o 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 vjf skill nn the, along with equivalent variations. In additian, the present invention is not c#escn-bed 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 Iangua~es are provided for disclosure of enablement and best mode of the present invention.
ja246] 'The present invention is well suited to a wide variety of computer network systems over numerous topologies. Within this field, the configuration and management of large networks comprise storage devices and computers that are communicatively coupled to dissimilar computers and storage devices over a network, such ash the Interned [0247] Finally, it should be noted that the language used in the specification has been pri'ncipaliy selected for readability and instructional purposes, and may not have been selected to delineate or circumscr~'be the inventive subject matter.
Accordingly, the disclosure of the present invention is intended to be illustrative, but not limiting, of the scope c~f the invention, which is set forth in the following claims.

Claims (5)

1. General topics 1. A method of automatically generating a description of a document, the method comprising:
retrieving a document in response to a query, the query comprising at least one query phrase;
determining for sentences of the document, a count of the at least one query phrase in the sentences;
selecting a plurality of sentences based on their respective phrase counts;
and forming a description of the document from the selected sentences.

2. The method of claim 1, wherein the count of the at least on query phrase comprises a first count, the method further comprising:
determining a second count of related phrases of the query phrase in a sentence;
determining a third count of phrase extensions of the query phrase in a sentence; and selecting a plurality of sentences based on their respective first, second and third counts.

3. The method of claim 1, wherein selecting a plurality of sentences based on their respective phrase counts, comprises:
sorting the sentences in declining order of their respective phrase counts;
and selecting a number of sentences having the highest phrase counts.
2. Personalized topics

4. A method of automatically generating a personalized description of a document, the method comprising:
storing a user model comprising a plurality of phrases contained in documents accessed by the user;

receiving a query from the user, the query comprising at least one query phrase;
selecting a document responsive to the query;
determining phrases that are related to the query and present in the user model; and generating a document description comprising selected sentences of the document, wherein the sentences are selected and ordered in the document description as a function of a number of the determined phrases in each sentence.

5. A method of automatically generating a personalized description of a document, the method comprising:
storing a user model comprising a plurality of phrases contained in documents accessed by the user;
receiving a query from the user, the query comprising at least one query phrase;
selecting a document responsive to the query;
determining phrases that are related to the query and present in the user model;
ordering the determined phrases wrath respect to the user model;
determining for sentences of the document, a count of each of the ordered phrases in the sentences;
selecting a plurality of sentences based on their respective phrase counts;
and forming a description of the document from the selected sentences.