CA2513852A1 - Phrase-based searching in an information retrieval system - 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 SEARCHING IN AN INFORMATION RETRIEVAL SYSTEM
Inventor: Anna L. PattQrson Cross 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/xxx,xxx, filed an July 26, 2004;
Phrase-Based Indexing in an Information Retrieval System, Application Na.1Q/xxx,xxx, filed on July 26, 2004;
Phrase-Based Personalization of Searches in an Information Retrieval System, Application No.1U/xxxxxx, filed on July 26, 2004;
Automatic Taxonomy Generation in Search Results Using Phrases, Application No.
10/x~oc,oxx, fried on July 26, 2004;
Phrase-Based Generation of Docmment Descriptions, Application No.10/xxxxxx, filed on July 26, 2004; and Phrase-Based 'Detection of Duplicate Docun:uents in an Information Retrieval System, Application No.10/xxx,x3cx, filed on July 26, 2004, all of which are co-owned, and incorporated by reference herein.
Field of the Invention I0002I The present invention relates to an information retrieval system for indexing, searching, and classifying documents in a large scale corpus, such as the Internet.
Background of the Invention [0003] Information retrieval systems, generally called search engines, are now an essential tool for finding information in large scale, diverse, and growing corpuses such as the Internet. Generally, search engines create an index that relates documents (or "page) to the individual words present in each document. A document is retrieved in response fio a query containing a number of query terms, typically based on having some:number of query terms present in the document The retrieved documents are then r, anked according to other sfiatistical measures, such as frequency of occurrence of the query terms, host domain, link analysis, and the Like. The retrieved documents are then presen#ed 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 docent is presented to provide the user with a glimpse of the document's content.
[00~~ 1?irect "Boolean" matching of query terms has well known limitations, and i,~ particular does not identify documents that do not have the query terms, but have related words. For example, in a typical Boolean system, a search on "Australian Shepherds" would not return d~uments about other herding dogs such as Border Collies that do not have the exact query terms. Rather, such a system is likely to also retrieve and highly rank documents that are about Australia (and have nothing to do with dogs), and documents about "shepherds" generally.
[OOUS~ The problem here is that conventional systems index documents based on iniiividual terms, than on concepts. Concepts are often expressed in phrases, such as "Australian Shepherd," "President of the United States," or "Sundance Film 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.
IndeaCing of phrases is typically avoided because of the perceived computational and memory requirements to identify all possible phrases of say three, four, or five or more words. For example, on the assumption that any five words could constitute a phrase,

2 and a large corpus would have at least 200,000 unique terms, there would approximately 32 x 10~ possible phrases, clearly more than any existing system could store'in memory or otherwise programmatically manipulate. A further problem is that phrases continually enter and Ieave the lexicon in terms of their usage, much more frequently than new individual words are invented. New phrases are ahNays being generated, from sources such technology, arts, world events, and law. Other phrases will decline in usage over time.
(00!16j Some existing information retrieval systems attempt to provide retrieval of concepts by using co~c:currence patterns of individual words. In these systems a search on o~ word, such as "President' will also retrieve documents that have other wordy that frequently appear with "President", such as "White" and " House."
While this approach may produce search results having documents chat are conceptually related at the level of individual words, it does not typically capture topical relationships that inhere between co-occurring phrases.
j0i)07J Accordingly, there is a need for an information retrieval systenn 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 phrases, and provide additional clustering and descriptive information about the documents.
Summary of the Invention (OOOg] An information retrieval system and methodology uses phrases to zndex, search, rank, and describe documents in the document collection. The system is adapted to identify phrases thathave sufficiently frequent and/or distinguished usage in the document collection to indicate that they are "valid" or "good"
phrases. In this manger multiple word phrases, for example phrases of four, five, or more terms, can be identified. This avoids the problem of having to identify and index every possible phra$es resulting from the all of the possible sequences of a given number of words.
[0009] The system is further adapted to identify phrases that are related to each otheF, based on a phrase's ability to predict the presence of other phrases in a document.
More specifically, a prediction measure is used that relates the actual co-occurrence rate of two phrases to an expected co-occurrence rate of the twfl phrases.
Information gain,, as the ratio of actual co-occurrence rate to expected co-occurrence rate, is one such prediction measure. Two phrases are related where the prediction measure exceeds a predetermined threshold. In that case, the second phrase has significant information gain tvith respect tv the first phrase. Semantically, related phrases will be those that are componly used to discuss or describe a given topic or concept, such as "President of the United States" and "White House." Fox a given phrase, the related phrases can be ordered according to their relevance.or significance based on their respective prediction measures.
[(g120j An information retrieval system indexes documents in the document collection by the valid or good phrases. For each phrase, a posting list identifies the documents that contain 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 phrase are also present in each document containing the given phrase. In this manner, the system can readily identify not only which documents contain which phrases in response to a search query, but which documents also contain phrases that are related to query phrases, and thus snore likely fn be specifically about the topics or concepts expressed in the query phrases.
(fall<) The use of phrases and related phrases further provides for the creation and use of clusters of related phrases, which represent semantically meaningful groupings of phrases. C~Sters are identified from related phrases That have very high prediction measure between all of the phrases in the cluster. Qusters can be used to organize the results of a search, including selecting which documents to include in the scarf h results and their order, as well as eliminating documents from the search results.
IOfll~j The information retrieval system is also adapted to use the phrases when searohirtg for documents in response to a query. The query is processed to identify any phrases 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 is in fact executed.
[OOT~) The related phrase information may also be used by the system to identify or select which documents to ixedude in the search result. The related phrase information indicates for a given phrase and a given document, which related phrases of the given phrase are present in the given document. Accordingly, far a query containing two query phrases, the posting list for a fixes 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 docunnents 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 separately process the posting list of the second query phrase, thereby providing faster search times. Of course, this approach may be expended to any number of phrases in a query, yielding in significant computational and timing savings.
(t7UI4j The system may be further adapted to use the phrase and related phrase information to rank documezits in a set of search results. The related phrase information of a given phrase is preferably stored in a format, such as a bit vector, which expresses 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 the bits are ordered according to the prediction measures {e.g., information gain) for the related phrases. The most significant bit of the related phrase bit vector are associated with the related phrase having the highest prediction measure, and the least sigxiificant 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 gas 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 Have low ordered related phrases. The bit vector value nnay also be used as a component in a more complex scoring function, and additionally may be weighted. The documents can then be ranked according to their document scores.
[00~5j Phrase information ntay also be used in an information retrieval system to personalize searches for a user. A user is modeled as a collection of phrases, for exanipIe, derived horn documents that the user has accessed (e.g., viewed on screen, prixtted, stored, etc.). More particularly, given a document accessed by user, the related phrases that are present in tins docum~erit, are included in a user model or profile.

lJurieng subsequent searches, the phrases in the user model are used to frlter the phrases of the search query and to weight the docutruent scores of the retrieved documents.
[001.x] Phrase information may also be used in an information retrieval system to create a description of a document, for example the documents included in a set of search results. 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 rnmbination), and some 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 whet the document is included in search results, so that the user obtains a better understanding of the document, relative to the query.
j(M1T~'j A further rat of this process of generating documextt descriptions allows the system to provide personalized descriptions, that reflect the interests of the user., A.s before, a user model stores information identifs~ing 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 common set is then ordexed 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 related phrases present in each document. A number of sentences having the highest number of common relahed phrases is selected as the personalized document description.

(ODl~] An information retrieval system may also use the phrase information to identify and eliminate duplicate documents, either while indexing (crawling) the document collection, or when processing a search query. Por a given document, each sentence of the document has a taunt of how many related phrases are present in the sentence. The sentences of document can be ranked by this counf, and a number of the top z~nlang, sentences (e.g., five sentences) are selected to form a document description.
This description is then stored in association with the document, fox example as a string or a dash of the sentences. During indexing, a newly crawled document is processed in the same 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 resuT~ts of a search query, the documents in the search result set can be processed to eliudnate duplicates.
[001] The present invention has further embodiments in system and software architectures, computer program products and computer implemented methods, and computer generated user interfaces and presentations.
(0020] The foregoing are just some of the features of an information retrieval systeu~ and methodology based on phrases. Those of skill in the art of information retriwal 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 Descrip~on of the Drawings j0021t] FIG. ~ is block diagram of the software architecture of one embodiment of the present invention.
[OU22j FIG. 2 illustrates a method of identifying phrases in documents.
[002g] FIG. 3 illustrates a document with a phrase windo~n,~ and a secondary window.
j0t12~J FIG. 4 illustrates a method of identifying related phrases.
(0025] FIG, b illustrates a method of indexing documents for related phrases, [0026] FIG. 6 illustrates a method of retrieving documents based on phrases.
j002'~J FIG. 7 iIlustrafies operations of the presentation system to present search results.
(402$] FIGS. 8a and 8b illustrate relationships between referencing and referenced documents.
(0021 The figures depict a preferred embodiment of the present invention for pur~ases of illustration only. One skilled in the art will readily recognize from the following discussion that alternative embodiments of the structures and methods illustrated herein may be employed without departing from the principles of the invention described herein.
Detailed Description of the Invention I. System Overview j403fl] Referring now to FIG. x, there is shown the software archifiecture of an embodiment of a search system 100 in accordance with one embodiment of present invention. In this embodiment, the system includes a indexing system 110, a search system 120, a presentation system 130, and a front end server 140.
jQ03~j The indexing system 110 is responsible for identifying phrases in documents, and indexing documents according to their phrases, by accessing various websites 190 and other document collections. The front end server 140 receives queries from a user of a client 170, and provides those queries ho the search system 120. The search system 120 is responsible for searching for 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 influence the ranking order. The search system 220 provides the search results to the presentation system 130. The presentation system 130 is responsible for modifyir~g the search results including removing near duplicate documents, and generating topical descriptions of documents, and providing the modified search results back to the front end server 140, which provides the results to the client 170. The system 100 further includes an index 150 that stores the indexing information pertaining to documents, and a phrase data store 160 that stores phrases, and related statistical inforn~ation.

[003] In the context of this application, "documents" are understood to be any type of media that can be indexed and retrieved by a search engine, including web 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 ox othei components, 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 Iimit~tion as to the scope of the invention is implied by the use of the generic term "documents." The search system 100 operates over a Iarge corp~xs of documents, such as the Internet and World Wide Web, but can likewise be used in more limited collections, such as 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 loss of generality then, the documents generally, regardless of format or location {e.g., which website or database) w01 be collectively referred to as a corpus or document colleciaon Each document has an associated identifier that uniquely identifies the document; the identifier is preferably a UItL,, but 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.
IL Indexing Svstem (0033] In one embodiment, the indexing system 110 provides three primary functional operations:1) identification of phrases and related phrases, 2) indexing of doc-~ments 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 T10 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 1T0 operates on an index 150 and data repository 160 of phrase data. These data repositories are further descrt-bed below.
I. Phrase Identification (0034] T'he phrase identification operation of the indexing system 110 identifies "goon" and "bad" phrases in the document collection that are useful to indexing and searcahing 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 indicted as having a distinguished appearance in such documents, such as delimited by marl~up tags or other morphological, format, or grammatical markers. Another aspect of good phrases is that they are predictive of other good phrases, and are not merely sequences of words that appear in the lexicon. Far example, the phrase "President of the United States" is a phrase that predicts other phrases such as "George Bush"
and "Bill Clinfion." However, other phrases are not predictive, such as "fell down the stairs" or "top of the morning," "out of the blue," since idioms and calloquisms like these tend to appear with many other different and unrelated phrases. Thus, the phrase identification phase determines which phrases are good phrases and which are bad {i.e., lacking in predictive power).
j0035] Referring to now FIG. 2, the phrase identification process has the following functional stages:
[0036] 2f14: Collect possible and good phrases, along with frequency and co-occurrence statistics of the phrases.
[003~j 202: Classify posss'ble phrases to either good or bad phrases based on frequency statistics.
[0038] 204: Prune good phrase list based on a predictive measure derived fxom the co-occurrence statistics.
j003'~J Each of these stages will now be described in fwrther detail 1004QJ The first stage 200 is a process by which the indexing system 110 crawls a set o~ documents in the document collection, making repeated partitions of the docuyent collection over time. One partition 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 uncxawled documents are processed in each partition, until all 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 T10 for each docqment that is crawled.
j004~) Traverse the words of ~e document with a phrase window length of n, where n is a desired maximum phrase length. The length of the wixtdow will typically be at least 2, and preferably 4 or 5 terms (words}. Preferably phrases include all words in the phrase window, including what would otherwise ire characterized as stop words, such:as "a", "the,' and so 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 30D during a traversal, showing the phrase window 3(?2 starting at the word "stock" and extending 5 words to the right.
The first word in the window 302 is candidate phrase i, and the each of the sequences i+T, r?+2, i+3, i+4, and i+$ ~ ~ewise a candidate phrase. Thus, in thzs example, the cand.~date phrases are: "stock", "stock dogsu, "stock dogs for", "stock dogs for the', "stock dogs for the Basque", and "stock dogs for the Basque shepherds".
[004~~ In each phrase window 302, each candidate phrase is checked in turn to determine if it is already present in the good phrase list 208 or the possible phrase list 206. ~f the candidate phrase is not present in either the good phrase list 20$
or the poss~le phrase list 206, 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 entzy gi, then the index 150 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 docu~nnent. An entry in the index 150 for a phrase gi (or a term) is referred to as the Case LOGO 13 posting list of the phrase g~. The posting list includes a list of documents d (hy their document identifiers, e.g. a document number, or alternatively a URL) in which the phra6e occurs.
(p04.5~ In addition, the co-occurrence matrix 212 is updated, as further explaiuned belo~r. In the very first pass, the good and bad lists will be empty, and thus, most phrases will tend to be added to the possible phrase list 206.
j0()~61 If the candidate phrase is not in the good phrase list 2D8 then it is added to the possible phrase list 206, unless it is akeady present therein Each entry p on the possible phrase list 206 has three associated counts:
[004~'J P{p): Number of documents an which the possible phrase appears;
jl'H»fd] 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 uinteresting" where the possible phrase is distinguished from neighboring content in the document by grammatical or format markers, for example by being in boldface, or underline, or as anchor text in a hyperlink, or in quotation marks.
These (and other) distinguishing appearances are indicated by various 1-fTML
markup language tags and grammatical markers. These statistics are maintained for a phrase when it is placed on the good phrase list 208.
[0050] In addition the various lists, a co-occurrence matrix ZIZ (G) for the good phza~es is maintained. The matrix G has a dimension of m x m, where m is the number of good phrases. Each entry G(j, k) in the matrix represents a pair of good phrases (g;, gk). The co-occurrence matrix 212 logically (though not necessarily physically) maitrtains three'separate counts for each pair (pi, gk) of good phrases with respect to a secondary windiow 304 that is centered at the current word l, and extends +/- h words.
7n one embodiment, such as illustrated in FIG. 3, the secondary window 304 is 30 words. The co-occurrence matrix 212 thus maintains:
(005] R(j,k): Raw Co-occurrence count. The number of times that phrase g;
appears in a secondary window 304 with phrase gx;
(005~j D(j,k): Disjunctive Interesting count, The number of times that either phrase g; or phrase gr~ appears as distinguished text in a secondary window;
and (0053] C(j,k): Corgunctive Interesting coun#: the number of times that both g;
and phrase gk appear as distinguished text in a secondary window. The use of the conjunctive interesting count is particularly beneficial to avoid the circumstance where a phrase (e.g., a copyright notice) appears frequently in sidebars, footers, or headers, and thus is not actually predictive of other text.
(005~j Referring to the example of FIG. 3, assume that the "stock dogs" is on the good phrase list 208, as well as the phrases "Australian Shepherd" and "Australian Shepard Qnb of America". Both of these latter phrases appear within the secondary window 304 around the current phrase "stock dogs". However, the phrase "Australian Shepherd Club of Anverica' appears as anchor text for a hyperlink (indicated by the undeprline) to website. Thus the raw co-occurrenre cocu~t for the pair {"stock dogs", "Australian Shepherd"} is incremented, and the raw occurrer~e count and the disju~ve interesting count for j"stock dogs', "Australian Shepherd Club of America"}
are both incremented because the latter appears as distinguished text.
(005$j The process of traversing each document with both the sequence window 302 arid the secondary window 304, is repeated for each document in the partition.
I005~1 Once the documents in the partition have been traversed, the next stage of the indexing operation is to update 202 the good phrase list 20$ from the possible phrase list 206. A possible phrase p on the possible phrase list 206 is moved to the good phrase list 208 if the frequency of appearance of the phrase and the number of doc~nents that the phrase appears in indicates that it has sufficient usage as semantically meaningful phrase.
[OQS~j In one embodiment, this is tested as folloyvs. A possible phrase p is removed fxom the possible phrase list 246 and placed on the good phrase list 208 if:
[005$] a) P(p) > 10 and S(p) > ZO (the number of documents containixtg phrase p is more than 14, and the number of occurxences of phrase p is more then 20};
or I0059j b) M(p) > 5 (the number of interesting instances of phrase p is more than 5) .
(006A] These thresholds are scaled by the number of documents in the pariition~
for example if 2,400,000 documents are crawled in a partition, then the thresholds are approximately doubled. Of coarse, those of skill in the art will appreciate that the specific values of the thresholds, or the logic of testing them, can be varied as desired.
(006L] Tf a phrase p does not qualify for the good phrase list 208, then it is checked for qualification for being a bad phrase. A phzase p is a bad phrase if:
j0062j a) number of documents containing phrase, P(pj < 2; and-j0063j b) number of interesting instances of phrase, M(p) = 0.
(OOC~] These conditions indicate that the phrase is both infrequent, and not used as indicative of significant corihent and again these thresholds rnay be scaled per number of documents in the partition.
(0065] It should be noted that the good phrase list 208 will naturally ira=lode individual words as phrases, in addition to mufti-word phrases, as desrn'bed above.
~6 This is because each the first word in the phrase window 302 is always a candidate phrase, 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) and multiple word phrases. The good phrase List 208 will also be considerably shori~er than the theoretical maximum based on all possible combinations of m phrases.
In typical embodiment, the good phrase list 208 will include about b.~x'105 phrases. A
list of bad phrases is not necessary to store, as the system need only keep track of possr~le and good phrases.
j0066j 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 large corpus. Thus, if say by the I0~ pass (e.g., 10,000,000 documents), a phrase appears for the very first tune, it is very unlikely to be a good phrase at that time. It may be new phrase just coming into usage, and thus during subsequent crawls becomes increasingly common. In that case, its respective counts will increases and may ultimately satisfy the thresholds for being a good phrase.
(OOG7~ The third stage of the indexing operation is to prune 204 the good phrase list 2Q8 using a predictive measure derived from the co-occurrence matrix 212.
Without pruning, the good phrase list 2U8 is likely to include many phrases that while legitimately 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 phrases. To identify good phrases, 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:

(OOti8) As noted above, the co-occurrence matrix 21~ is an m x m matrix of storixig data associated with the good phrases. Each row j in the matrix represents a good phrase g; and each column k represented a good phrase gk. For each good phrase gj, an expected value E(g;) is computed. The expected value E is the.
percentage of docuineni~ in the collection expected to contain g~. This is computed, for example, as the ratio:of the number of documents containing g; to the total number T of docu:auents in the collection that have 'been crawled: P~j)/T.
(006~j As noted above, the number of documents contairUng g; is updated eactl time g; appears in a document. The value for E(g~) can be updated each time the counts for gi are incremented, or during this third stage.
I007i)j Next, for each other good phrase gk. (e.g., the columns of the matrix), it is detiermined whether g; predicts gx. A predictive measure for g, is determined as follows:
(007x] i) compute the expected value E(gk). The expected co-occurrence rate E(j,k~ of g; and g~, if they were unrelated phrases is then E(g;)*E(gx);
(00721 ii) compute the actual co-occurrence rate A(j,k) of g; and gk. This is the raw co-occurrence count R(j, k) divided by T, the total number of documents;
(0073] iii) g; is said to predict gk where the actual co-occurrence rate A~,lc) exceeds the expected co-occurrence rate E{jk) by a threshold amount.
(0074] In one embodiment, the predictive measure is information gain. Thus, a phrase g; predicts another phrase g~ when the information gain I of gx in the presence of ga e~cc~eds a threshold. In ane embodiment, this is computed as follows:
(0(J75] I(~.k) = A(j,k)/E(j,k) (007G] And good phrase g; predicts good phrase gt where:

[0(y79~ I(~,kj > Information Gain threshold.
[007$] In one embodiment, the infarrnation gain threshold is 1.5, but is preferably between 1.I and 2.7. Raising the threshold over 1.0 serves to reduce the possjbility that two otherwise unrelated phrases co-occur more than randomly predicted.
Iou7g) 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 plu~ases g~ exceeds the information gain threshold, then Khis means that phrase g; does not predict any other good phrase. In that case, g; is removed from the good phrase list 208, essentially becoming a bad phrase. Note that the cblumn j for the phrase g~ is not removed, as this phrase itself may be predicted by other good phrases.
[f)08~] This step is concluded when all rows of the co-occurrence matrix 2'I2 have~been evaluated.
j0081] 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 predichs its phrase extensions, and which starts at the left most side of the phrase (i.e., the begirming of the phrase). The "phrase extension" of phrase p is a super-~sequence that begins witi~ phrase p. For example; the phrase °President of" predicts "President of the United States", "President of Mexico", "President of AT&T", etc. All of these loiter phrases are phrase extensions of the phrase President of since they begin with President of" and are super sequences thereof.
10082) Accordingly, each phrase g~ remaining on the good phrase list 208 will predict some number of other phrases, based on the information gain threshold previously discussed. Now, for each phrase g~ the indexing system 120 performs a string match with each of the phrases gx that is predicts. The string ma6ch tests whether each predicted phrase gx is a phrase extension of the phrase g;. If all of the predicted phrases gx are phrase extensions of phrase g;, then phrase g~ is incomplete, and is removed from the good phrase List 208, and added to an incomplete phrase list 216.
Thus, ~f there is at least one phrase gx that is not an extension of g;, then g;is complete, and maintained in the good phrase list 20~. For example then, "President of the United" is an incomplete phrase because the only other phrase that it predicts is "President of the United States" which is an extension of the phrase.
IOOSSj The incomplete phrase list 216 itself is very useful during actual seanhing: When a search query is received, it can be compared against the inoompIete phase list 216. If the query (or a portion thereof) matches an entry in the list, then the seardh system 220 can lookup the most likely phrase extensions of the incomplete phrase (the phrase extension having the highest information gain given the incomplete phrase).
and suggest this phrase extension to the user, or aufiamatirally search on the phrase extension. For example, if the search query is "President of the United," the search system 120 can automatically suggest to the user "President of the United States" as the search query.
~008~j After the last stage of the indexing process is completed, the good phrase list 208 will contain a large number of good phrases that have bean discovered in the corps. Each of these good 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 freduency and independence to represent meaningful concepts 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. Further, since the above process of crawling and indexing is repeated periodically as new documents are added to the document collection, the indexing system IIO automatically detects new phrases as they enter the lexicon.
Z. ~ldentification of Related Phrases and Clusters of Related Phrases j008,~] Referring to FIG. 4, the related phrase identification process includes the following functional operations.
jOt?Siy1 400: Identify related phrases having a high information gain value.
[fl08yJ 402: Identify clusters of related phrases.
[0088] 404: Store cluster bit vector and cluster number.
j0089j Each of these operations is now described in detail.
j0U9~] First, recall that the co-occurrence matrix 212 contains good phrases g;, each:of which predicts at least one other good phrase gx with an information gain greater than the information gain threshold. To identify 400 related phrases then, for each pair of good phrases (g;, gx) the information gain is compared with a Related Phrase threshold, e.g.,100. That is, g; and gx are relafied phrases where:
(~j 1~8~~ ~) ~ 1~.
[009] This high threshold is used to identify the co-occurrences of good phrases that are well beyond the statistically expeched rates. Statistically, it means that phrases g;
and gx co-occur 100 times more than the expected co-occurrence rate. For example, given the phrase "Monica Lewinsky" in a document, the phrase °BiII
Clinton" is a ItlO
times more likely to appear in the same document, then the phrase "Bill Clinton" is likely to appear on any randomly selected document. Another way of saying this is that the accuracy of the predication is 99.999°6 because the occurrence rate is 100:1.

(009tij Accordingly, any entry (g,;, g~) that is less the Related Phrase threshold is zeroed out, indicating that the phrases g;, gk are not related. Any rennaining entries in the oo-occurrence matrix 2I2 now indicate alI related phrases.
I009#] The columns gk in each raw g~ of the co-occurrence matrix 212 are then sorted by the inforrnatian gain values I(gJ, g~}, so that the related phrase gx with the highest infoxznation gain is listed first. This sorting thus identifies for a given phrase g,;, which oilier phrases are most likely related in terms of information gain.
[0095] The next step is to determine 402 which related phrases together form a cluster of related phrases. A cluster is a set of related phrases in which each phrase has high!information gain with respect to at least one other phrase. In one embodime?t~, clusters are identified as follows.
(00~] 1n each row g; of the matrix, there will be one or more other phrases that are related to phrase gf. This set is related phrase set Ry where R~{g,~ gr, ...g,~).
(009~j For each related phrase m in R~, the indexing system 110 detemiic~es if eachof the other related phrases in R is also related to g;. Thus, if 1(gk, gi} is also non-zero,j then gr, gk, and gi are part of a cluster. This cluster test is repeated far each pair (g~, gm} i~ R.
(009$] For example, assume the good phrase "Bill Qiriton" is related to the phrases "President", "Monica Lewinsky", because the information gain of each of these phrases with respect to "Bill Clinton' exceeds the Related Phrase threshold.
Further assume that the phrase "Monica Lewinsk~' is related to the phrase "purse designer".
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 determnni~ng their corresponding information gains. Thus, the indexing system 110 determines the information gain I("President", "Monica Lewinsky"), I("President", "purse designer', and ~o for#h,, for all paixs in R. In this example, "Bill Ciinton,"
"President", and "Monica Lewinsky" form a one cluster, "Bill Ciinton," and "President" farm a second cluster, and "Moruca Lewinsky " and "purse designer' form a third cluster, and "Monica Lewinsky", "Bill Clinton," and "purse designer" form a fourth cluster. This is because while "Bill C~inton" does not predict "purse designer" with sufficient information gain, "Monica Lewinsky" does predict both of these phrases.
(j To record 404 the duster information, each cluster is assigned a unique dusiteer number (cluster 1D). This information is then recorded in conjunction with each good phrase g-.
j0I0~j In one embodiment, the cluster number is determined by a cluster bit vector that also indicates the orthogonality relationships between the phrases. The duster bit vector is a sequence of bits of length n, the number of good phrases in the good phrase list 208. For a given good phrase g;, the bit positions correspond to the sorted related phrases R of g;. A bit is set if the related phrase gx in R is in the same cluster as phrase gJ. More generally, this means that the corresponding bit in the cluster bit victor is set if there is information gain in either direction between g;
and gf.
(07.01 j The cluster number then is the value of the bit string that results.
This impl~nentation has the property that related phrases that have multiple or one-way information gainappear in the same cluster.
(010Zj An example of the duster bit vectors are as follows, using the above phrases:
Monica purse Quster Bill Clinton President Lewinsky designer ID

Bib Clinton1 1 1 0 I4 President 1 1 0 0 12 Mbnica Le~lvinskv1 0 1 1 1I

purse deli er fl 0 I I 3 (4I0~~ 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;, gk) from highest to lowest. In addition, fox each good phrase g;, there will be a clustier bit vector, the value of which is a cluster number identifying the primacy chxster of which the phrase g~ is a member, and the orthogonality values (1 or 0 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 "Moni~ca Lewinsky"
are in cluster 14 based on the values of the bits in the row fox phrase "Bill Clinton".
[Ol.~j To store this information" two basic representations are available.
First, as indicated above, the information may be stored in the co-occurrence matrix 212, wherein:
(01I?~~ entry G[row j, col. k] ~ (I(j,k), clusterNumber, clusterBitVector) (U10~) Alternatively, the matrix representation can be avoided, and all information stored in the good phrase list 208, wherein each row therein represents a good Phrase ~:
(020~j Phrase rows = list (phrase gk, (I{j,k), clusterNumber, clusterBitVector)~.
[0101 This approach provides a useful organization for clusters. First, rather than ~ strictly-and often arbitrarily-defined hierarchy of topics and concepts, this apprbach recognizes that topics, as indicated by related phrases, form a complex graph of r~llataonships, where some phrases are related to many other phrases, and some phraises have 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 characterized "local" to each good phrase, and some dusters will then overlap by having one or more common related phrases.
[010$j For a given good phrase g; then the ordering of the related phrases by infornnation 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 sign~cant phrases that appear in the document coltectian, and beneficially, the way these related phra_~es are used together in natural "clusters" in actual practice. As a result, this data-driven clustering of related phrases avoids the biases that are inherent in any manually directed "editorial" selection of related terms and concepts, as is common in many systenns.
3. Indexing Documents with Phrases and Related Phrases j01T1j Given the good phrase list 208, including the information pertaining to related phrases and clusters, the next functional operation of the indexing system lI0 is to index documents in the document collection with respect to the good phrases and clusters, and store the updated information in the index I50. FIG. S
illustrates this process, in which #here are the following functional stages for indexing a document;
(0112 5~: Post document to the posting lists of good phrases found in the document.

[0113) 502: Update instance counts and related phrase bit vector fox related phases and secondary related phrases.
[011i~J 504: Annotate documents with related phrase information.
[p11~) 50b: Reorder index entries according to posting list sizx.
[0~1~J These stages are now described in further detail.
[pll:J~j A set of documents is traversed or crawled, as befoxe; this may be the same or a different set of documents, For a given document d, traverse 500 the document word by word with a sequetuce window 302 of length n, from-position i, nn the manner described above.
j0~1$j In a given phrase window 302, identify all good phrases in the window, starting 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.
[pll~J For each good phrase g; (example g1 "President" and g4 "President of ATT"} post 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 do~nent.
[412(1] In one embodiment, the posting list for a phrase g; takes the following logical form:
j0121J Phrase g;: list: (document d, [list related phase counts] [related phrase information]) [p12~] For each phrase g~ there is a list of the documents d on which the phrase appears, For each document, there is a list of counts of the number of occurrences of the reIatea phrases R of phrase g; that also appear in document d.

(OI~'i] In one embodiment, the related phrase information is a related phase bit vectpr. This bit vector may be characterized as a "bi-bid' vector, in that for each related phrase gx there are two bit positions, g~-1, gx2. The first bit position stores a flag indicating whether the related phrase gF is present in the document d (i.e., the count for gx in uocument d is greater than 0). The second bit position stores a flag that indicates whether a related phrase gr of gx is also present in document d. The related -phrases gr of a related phrase gk of a phrase g~ are herein called the "secondary related phrases of g; "
1fie counts and bit positions correspond to the canonical order of the phxases in R
(sorted in order of decreasing information gain). This sort order has the effect of making the related phrase gx that is mast highly predicted by g,; associated with the most significant bit of the related phrase bit vector, and the related phrase gr that is least predicted by g; associated with the least significant bit.
Ial2~] It is useful to note that far a given phrase g, the length of the related phrase bit vector, and the association of the related phrases to the individual bits of the vector, will be the same with respect to all doc~unents 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 givien related phrase. This is beneficial for facilitating the search process i~o identify documents in response to a search query. Accordingly, a given docunnent will appear in the posting lists of many different phrases, and in each such posting list, the related phra$e vector for that document will be specific to the phrase that owns the posting list.
This aspect preserves the locality of the related phrase bit vectors with respect to indi~dual phrases and documents.
z~

~012~j Accordingly, the next stage 502 includes traversing the secondary window 304 of the current index position in the document (as before a secondary window of +/- K terms, for example, 30 terms), for example from i-K to i+IC
For each related phrase gx of g; that appears in the secondary window 304, the indexing system 110 itzcrements the count of gk with respect to document d in the related phrase count. If g; appears Later in the document, and the related phrase is found again within the Iafier secoxtdary window, again the count is incremented.
~012~ j As noted, the corresponding first bit gk-1 in the related phrase bit map is set based on the count, with the bit set to 1 if the count for gk is >0, or set tao 0 if the count equals 0.
j012Tj Next, the second bit, grr2 is set by looking up related phrase gk in the index 150, identifying in gx's posting list the entry for document d, and then checlung the seconuary 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.
(0I2$) When document d has been completely processed in this rnarmer, the indexing system 120 will have identified the following:
(0I29) i) each good phrase g; in docunnent d;
j013!7] ii) for each good phrase g~ which of its related phrases g~ are present in docufnent d;
X0131) iii) for each related phrase gk present in document d, which of its related phrases g (the secondary related phrases of gJ} are also present in document d.

a) Deterzninin~ the Topics for a Document (013'Z1 The indexing of documents by phrases and use of the clustering information provides yet another advantage of the indexing system 1I0, which is the abTliiy to determine the topics that a document is about based on the related phrase information.
I013~~ Assume that for a given good phrase g; and a given document d, the posting list entry is as follows:
(413~~ gi: d~ument d: related phrase counts= (3,4,3,0,0,2,1,1,0}
[pI35~ related phrase bit vector:={ 111110 00 00101010 OT}
j0iwhere, the related phrase bit vector is shown in the bi-bit pairs.
(013] From the related phrase bit vector, we can determine primary and secondary topics for the document d. A primary topic is indicated by a bit pair (l,Tj, and a secondary topic is indicated by a bit pair (1,0j. 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 seccmdary related phrases gc as well This may be interpreted to mean that the author of the document d used several related phrases g~, g~, and gc together in drafting the c~oc.-ument. A bit pair of (1,0j indicates that both ga and gt are present, but no further secondary related phrases from gt are present, and thus this is a less significant topic.
b) Document Annotation for Irn~roved Ranlane (0138] A further aspect of the indexing system 110 is the ability to annotate each document d during the indexing process with information that provides for imprbved ranking during subsequent searches. The annotation process 506 is as follows.

A given document d in the document collection may have some number of outlinks to other documents. Each outIink (a hyperlink) includes anchor text and the document identifier of the target document. For purposes of explanation,, a current docw'ment d being processed will be referred to as URLO, and the target document of an outlihk on document d will be referred to as URL1. For later use in ranking docunnents in ss~asch results, far every Iink in URLO, which points to some other URLi, the indexing system 110 creates an outiink score for the anchor phrase of that link with respect to IJRLp, and an iniink score for that anchor phrase with respect to URLi. That is, each link xn the document collection has a pair of scores, an outlink score and an inlink score.
These scores are computed as follows.
j014Q] On a given document URLO, the indexing system 110 identifies eadt outlii~k to another document URL2, in which the anchor text A is a phrase in the good phrase list 208. FIG. 8a illustrates schematically this relationship, in which anchor text "A" ire document URLO is used in a hyperlink 800.
(02411 In the posting list fox phrase A, URLO is posted as an outlink of phrase A, and I'TRL1 is posted as an inlink of phrase A. For URLO, the related phrase bit vector is rnmRleted as described above, to identify the related phrases and secondary related phrases of A present in URLO. This related phrase bit vector is used as the outiink score for ttte link from URLO to URL1 containing anchor phrase A.
(0142] Next, the iniink score is determined as follow. For each inlink to URL1 containing the anchor phrase A, the indexing system 110 scans URL1, and determines whether phrase A appears in the body of URLl. if phrase A not only points to URLl (via a outlink on URLO), but also appears in the content of URL1 itself, this suggests that URLl can be said to be intentionally related to the concept represented by phrase A.

FIG. ~Bb illustrates this case, where phrase A appears in both URLO (as anchor text) and in thg body of URLl. In this case, the related phrase bit vector for phrase A
for URLl is as the inlink score for the link from URLO to URLI containing phrase A.
If the anchor phrase A does not appear in the body of URL1 (as in FIG.
a different step is taken to detezniine the inlink score. In this case, the system 120 creates a related phrase bit vector for URLI for phrase A (as if A rwas present in URLI) and indicating which of the related phrases of phrase A
in URL1. This related phrase bit vector is then used as the inlink score far the URLO to URL1.
For example, assume the foDowing phrases are initially present in URLO
and ~,JRL1:
Anchor Phrase Related Phrase Bit Vector Australian blue red agility Document She herd Aussiemerle merle tricolortr ' URL p 1 1 0 U 0 0 uRL l 1 0 I ~ I 0 [OZ ) ~ (In the above, and following tables, the secondary related phrase bits are not ~own~.
s The URLO
row is the outlink score of the link from anchor text A, and the URL~
row is the inlink score of the link.
Here, URLO
contains the anchor phrase "Australian Shepard"
which targets URL1.
Of the five related phrases of "Australian Shepard", only one, "Aussie"
appears in URLO.
Intuitively then, URLfl is only weakly t abou Australian Shepherds.
URL1, by comparison, not only has the phrase "Australian Shepherd"
pxesent in the body of the document, but also has many of the related I

phra;;es present as well, "blue merle,"
"red merle,"
and "tricolor.
Accordingly, beta ~
se the anchor phrase "Australian Shepard"
appears in both ITRLO
and URL1, the score for URLO, and the inlink score for URL1 are the respective rows shown a (~ 4~~ The second case described above is inhere anchox phrase A does not in URLI. To that, the indexing system X10 scans URLI and determines which of the related phrases "Aussie," "'blue merle," "red merle," "tricolor;' and "agility are present in URL2, and creates an related phrase bit vector accordingly, for Anchor Phrase Related Phrase Bit Vector Australian blue red agility T~ocument She herd Aussiemerle merle tricolortrainin URLD _ 1 1 0 0 0 0 iJRL~ 0 0 x 1 2 0 [014 Here, this shows that the URL2 does not contain the anchor phrase I
~'Au~tralian Shepard", but does contain the related phrases "blue merle', "red merle", and i tricolor".
(42 ~ J 'This approach has the benefit of entirely preventing certain types of manipulations of web pages (a class of docun~ertts} in order to skew the results of a sear . Search engines chat use a ranking algorithm that relies on the number of links that mt to a given document in order to rank that document can be Nbombed" by artificially creating a large number of pages with a given anchor text which then point to~
a de aired page. As a result, when a search query using the anchor text is entered, the I
desi ied page is typically returned, even if in fact this page has little or nothing to do withlthe anchor text. Importing the related bit vector from a target document 1JRL1 into the phase A related phrase bit vector for document URLO eliminates the reliance of the search system on just the relationship of phrase A in URLO pointing to IJRL2 as an indicator of significance or URL1 to the anchor text phrase.
j0149j Each phrase in the index 250 is also given a phrase number, based on its I
freq f ency of occurrence in the corpus. The more common the phrase, the Iower phrase number it receivesorder in the index. The indexing system 110 then sorts 506 all of the I
pos ~ g lists in the index 150 in declining order according to the number of documents liste ~ phxase number of in each posting Iist, so that the most frequently occurnng phra~ are listed first. The phrase number can then be used to look up a particular phrase.
f III. Search System [0150] The search system 220 operates to receive a query and search for documents relevant to the query, and provide a List of these documents with links to tire documents) in a set of search results. FIG. 6 illustrates the main functional operations of the search system 120:
[0151j 600: Identify phrases in the query.
[015 ~ j 602: Retrieve documents relevant to query phrases.
[0I5 ~ j 604: Rank documents in search results according to phrases.
jai5~j The details of each of these of these stages is as follows.
1. Identification of Phrases in the duer""y and Queryr Expansion 1015 j The first stage 600 of the search system 120 is to identify any phrases that are went in the query in order to effectively search the index. The following terminology is used in this section [015isj , q: a query as input and receive by the search system 120.
j015~j Qp: phrases present in the query.

Qr: related phrases of Qp.
Qe: phrase extensions of Qp.
Q: the union of Qp and L,~x.
A query q is received from a client 190, having up to some ma~dmum of characters or words.
A phrase window of size N (e.g., 5} is used by the search system 120 to the terms of the query q. The phrase window starts with the first term of the xtends N terms to the right. This window is then shifted right M N times, Z is the number of terms in the query.
J
[016 ~ ) At each window position, there will be N terms (or fewer) t~,~nns in the window. These terms constitute a possible query phrase. The possible phrase is looked up irf the good phrase Iist 208 to determine if if is a good phrase or not. If the possible phrase is present in the good phrase List 208, then a phrase nuanber is returned for phrase; the possible phrase is now a candidate phrase.
j13T6~j After all possible phrases in each window have been tested too deb if trey are good candidate phrases, the search system 120 will have a set of phrase numbers for the corresponding phrases in the query. These phrase numbers are then Sor ~d {declining order).
[016] Starting with the highest phrase number as the first candidate phrase, the sear system 120 determines if there is another candidate phrase within a fixed numerical distance within the sorted list, i.e., the difference between the phrase numbers is within a threshold amount, e.g. 20,000. If so, then the phrase that is leftmost in the t query is selected as a valid query phrase Qp. This query phrase and all of its sub-phrases is removed 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.
jp166~ For example, assume the search query is "Hillary Rodham C7inton Bill on the S~be Floor". The search system 120 would identify the following candidate phrases, "1-ii>lary Rodham Clintnn BMl on," "Hillary Rodham Clinton Bill,' and "Hillary Rodliam Clinton". The first two are discarded, and the last one is kept as a valid query I
phra~. Next the search system 120 would identify "Bill on the Senate Floor", and the subs~hrases "Br71 on the Senate", °BiII on the', "Bill on", "Bill", and would select "Bill"
I
as a valid query phrase Qp. Finally, the search system 120 would parse "on the senate floor"' and identify uSenate Floor" as a valid query phrase.
[016 l Next, the search system 120 adJttsts the valid phrases Qp for capithlization When parsing the query, the search system 120 identifies potential capitalizations in each valid phrase. This may be done using a table of known capitalizations, such as "united states" being capitalized as "United Stafies", or by using a grammar based capitalization algorithm. This produces a set of properly capitalized query phrases.
[016] The search system 120 then makes a second pass through the capitalized phra$es, and selects only those phrases are leftmost and capitalized where both a phrase and ~~s subphrase is present in the set. For example, a search on "president of the uni Id states" will be capitalized as "President of the United States".
(016~j In the next stage, the search system 120 identifies 602 the documents that are relevant to the query phrases Q. The search system 120 then retrieves the posting lisfs ~ f the query phrases Q and intersects these lists to determine which documents appear on the all (or some number) 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 sear ' system 120 first forms the union of the posting lists of the phrase exten~io~ns, prior to dqulg the intersection with the posting lists. The search system 120 identifies phrase extensions by looking up each query phrase Q in the incomplete phrase list 216, as desc~i~ibed above.
[817~J The, result of the intersection is a set of documents that are relevant to the query. Indexing doclunents by phrases and related phrases, identifying phrases Q in the query, and then expanding the query to include phrase extensions results in the selection of a set of documents that are more relevant to the query then would result in a conventional Boolean based search system in which only documents that contain the quern terms are selected.
[0171) Tn one embodiment, the search system 120 can use an optimized mechanism to identify documents responsive to the query without having to intersect all of the posting lists of the query phrases Q. As a result of the structure of the index 150, fox a ch phrase g;, the related phrases gx are kno~nrn and identified in the related phrase bit vector far gk. Accordingly, t~~is 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. In those cases, the related phrase bit vectors can be directly acce used, and then used next to retrieve corresponding documents. This pnxess is mor fully described as follows.
j0171 J Given any two query phrases Q1 and Q2, there are three possible cases of relations:
j027~J 1) Q2 is a related phrase of QI;

(0T7 j 2) Q2 is not a related phrase of Q1 and their respective related phrases Qr1 d Qr2 do not intersect (ie., no common related phrasesj; and jUl.7 j 3) Q2 is not a zelated phrase of Q1, but their respective related phrases Qr1 ~nd Qr2 do intersect.
For each pair of query phrases the search system X20 determines the case by looking up the related phrase bit vector of the query phrases Qp.
The search system 224 proceeds by retrieving the posting list for query Q1, which contains the documents containing Q1, and for each of these documents, a related phrase bit vector. The related phrase bit vector for Q1 will indicated whether phrase Q2 (and each of the remaining query phrases, if any) is a rely ~ d phrase of Q2 and is present irt the document [017$j if the first case applies to Q2, the search system 120 scans the related phra$e bit vector for each document d in Ql's posting list to determine if it has a bit set for Q2. if this bit is not ~t in for document d in Q1's posting list, then it means that Q2 does i~ot appear in that document. As result, this document can be immediately I
eliminated from further consideration. The remaining documents can then be scored.
This ~ eans further that it is unnecessary for the search system 120 to pxocess the posting lists ~ f Q2 to see which documents it is present in as well, thereby saving compute time.
jDI79j If the second case applies to Q2, then the two phrases are unrelated to each ether. For example the query "cheap bolt action rifle" has two phrases "cheap" and "boI action rifle". Neither of these phrases is related to each other, and further the rela ~ phrases of each of these do not overlap; i.e., "cheap" has related phrases "low COSt~ ~ "IIleXpeI4SiVe," "dlsCOUnt, ~ "bargalll basement," arid "lousy,", whereas "bolt action rifle" has related phrases "guru," ~'22 caliber', "magazine fed,' and "Armalite AR-30M'i, which lists thus do not intersect. In this case, the search system 120 does the regular intersection of the posting lists of Q1 and Q2 to obtain the documents for scoring.
t j01$ ~ J 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 "bal ~ action rifle" and "22" would both have "gun" as a related phase. In this case, the search system 120 retrieves the posting lists of both phrases Ql and Q2 and intersects the fiats to produce a list of documents that contain both phrases.
The search system 120 can then quickly score each of the resulting First, the search sysfiem 120 determines a score adjustment value for each The score adjustment value is a mask formed from the bits in the positions ing to the query phrases Q1 and QZ in the related phrase bit vector for a document. For example, assume that Q1 and C22 correspond to the 3~ and 6~ bi-bit in the related phrase bit vector for document d, and the bit values in 31'd are {1,I) and the bit values in the b~ pair are {1,0~, then the score adjustment is the bit mask "00 X012 00 0010". The scare adjustment value is then used to the related phrase bit vector for the documents, and modified phrase bit vectors then are passed into the ranking function {next described) to be used in calculating a score for the documents.
2. Rankine a) Ranking Documents Based on Contained Phrases (018?] The search system 120 provides a ranking stage 604 in which the d ~ enrs in the search results are ranked, using the phrase information in each doculnen~s related phrase bit vector, and the cluster bit vector for the query phrases.

This i pproach ranks documents according to the phrases that are contained in the document, or informally "body hits:' [Ol$~j As described above, for any given phrase g;, each document d in the gr's I
posting iist has an associated related phrase bit vector that identifies which related phrases g~ and which secondary related phrases gr are present in document d.
The more relatkd phrases and secondary related phrases present in a given document, the more bits . t wi71 be set in the document's related phrase bit vector for the given phrase. The mo ~ its that are set, the greater the numerical value of the related phrase bit vector.
(OTB~j Accordingly, in one embodiment, the search system 120 sorts the documents in the search results according to the value of their related phrase bit vectors.
The dl ocuments containing the most related phrases to the query phrases ø
will have the high ~ t valued related phrase bit vectors, and these docwments will be the highest ranking documents in the search results.
j(17.85] This approach is desirable because semantically, these dots are most topically relevant to the query phrases. Note that this approach provides highly relevl t documents even if the documents do not contain a high frequency of the input query terms q, since related phrase information was used to both identify relevant docu~ ents, and theat rank these documents. Documents with a iow frequency of the input query terms may still 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 the query terms and phrases but no related phrases.
j0I86j In a second embodiment, the search system 120 scores each document in the result set according which related phrases of the query phrase Q it contains. This is donelas follows:
i I
[018 ~ Given each query phrase Q, there will be some number N of related phrases Qr to the query phrase, as identified during the phrase identificaiion process.
As described move, the related query phrases Qr are ordered according to their I
information gain from the query phrase Q. These related phrases are then assigned poin ~ , started with N points for the first related phrase Qr1 (i.e., the related phrase Qr I
with the highest information gain from Q), then N-2 points for the next xelated phrase Qr2, ~ en N-2 points for Qr3, and so on, so that the last related phrase QrN
is assigned 1 P~~.
r [~1$$j Each document in the search results is then scored by determining which rela ~ phrases Qr of the query phrase Q are present, and giving the document the pom~s assigned to each such related phrase Qr. The documents are then sorted from highest to lowest score.
[fl1$~j As a further refinement, the search systean 220 can cull cerfain documents result set. In some cases documents may be about many different topics; this is the case for longer documents. In many cases, users prefer documents that on point with respect to a single topic expressed in the query over that are relevant to many different topics.
To cull these latter types of documents, the search system 120 uses. the clustør information in the cluster bit vectors of the query phrases, and removes any document in which there are more than a threshold number of clusters in the dot.
For ~ ample, the search system 12~ can remove any documents that contain more than two etusters. This cluster threshold can be predeed, or set by the user as a search para eter.

b) Ranking Documents Based on Anchor Phrases ~4I9)~~ In addition to ranking the documents in the search results based on body hits if query phrases Q, in one embodiment, the search system 120 also ranks the docu~ is based on the appearance of query phrases Q and related query phrases Qr in anchors to other docwnents. In one embodiment, the search system 120 calculates a ach document that is a function (e.g., linear combination) of two scores, a :ore and an anchor hit score.
For example, the document score for a given document can be calculated Score = .30*(body hit score)+.70*(anchor hit score}, 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 vechor for the document, given the query phrases Qp, in the manner described above.
this value can directly obtained by the search system 12(T by looking up each cjuery phrase f~ in the index 150, accessing the document from the posting list of the query phrase Q and then accessing the related phrase bit vector.
The anchor hit score of a document d a function of the related phrase bit of the query phrases Q, where Q is an anchor term in a document that references ant d. YYhzn the indexing system 110 indexes the documents in the document ~n< it maintains for each phrase a list of the documents in which the phrase is text in an outlink, and also for each document a list of the inlinks (and the ed anchor text) from other documents. The iniinks for a document are references (e.g. ~yperlinks) from other documents (referenceng documents) fio a given document.

ral9 ~ J To determine the anchor hit score for a given document d then, the search sys 120 iterates over the set of referencing documents R (i=1 to number of refer ncutg documents) listed in index by their anchor phrases Q, and Burns the follolvmg product:
r [0199) R;.Q.Related phrase bit vector*D.Q.Related phrase bit vector.
The product value here is a score of how topical anchor phrase Q is to D. This score is here called the "inbound score component." This product weights the current document D's related bit vector by the related bit vectors of anr=hoc phrases in the referencing document R If the referencing documents R
are related to the query phrase Q (and thus, have a higher valued related bit vector), then this increases the significance of the current document D
score.
The body hit score and the anchor hit score are then combined to aeate the document as described above.
Next, for each of the referen~cirtg documents R, the related phrase bit for each anchor phrase Q is obtained. This is a measure of how topical the anchor Q is to the document R. This value is here called the outbound scare component.
From the index.150 then, all of the (referencing document, referenced pairs are extracted for the anchor phrases Q. Thess pairs are then sorted by their ~sssoci,ated (outbound score component, inbound score component) values.
on the implementation, either of these components can be the primary sort key, ~nd the other can be the secondary sort key. The sorted results are then presented to th~ user. Sorting the documents on the outbound score component makes documents that 1?ave many related phrases to the query as anchor hits, rank most highly, thus these documents as "expert" documents. Sorting on the inbound score makes documents that frequently referenced by the anchor terms the ranked.
3. Phrase Based Personalization of Search Another aspect of the search system X20 is the capability to personalize G06 o~r customize the ranking of the search results in accordance with a model of the particular interests. In this maiuier, documents that more likely to be relevant to the user's interests are ranked higher in the search results. The personalization of search is as follows.
As a preliminary matter, it is useful to define a user's interests (e.g., a user in terms of queries and documents, both of which can be represented by phrases.
For a~ input search query, a query is represented by the query phrases Q the related phrases of Qr, and phrase extensio~ts 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 repr anted by the phrases associated with the page. As described above, given a query and c3ocum~ent, the relevant phrases for the document are determined from the body sco ~ (the related bit vectors) for all phrases indexed to the document Finally, a user can ~ represented as the union of a set of queries with a set of documents, in terms of the phases that represent each of these elements. The particular documents to include in th~ set representing the user can be determined from which documents the user seleyn 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 dest~'I ations.
0203] The process of constructing and using the user model for personalized ra ~ g is as follows.

j02 ,) First, for a given user, a list of the Iast K queries and P documents accessed is maintained, where K and P are preferably about 254 each. The lists may be I
in a user account database, where a user is recognized by a Iogin or by cookies. For a given user, the lists ~~iIl be empty the first time the user provides Next, a query q is received from the user. The related phrases Qr of q are along with the phrase extensions, in the manner described above. 'This forms model.
In a first pass (e.g., if there are no stored query information for the usery, the search system 124 operates in simply return the relevant documents in the search result to the user's .query, without further customized ranking.
j024?j A client side browser tool monitors which of the documents in the search resuIjs the user accesses, e.g., by clicking on the document link in the search results.
Thes ~ accessed documents for the basis for selecting which phrases will become part of the user model. For each such accessed document, the search 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.
Next, given the phrases related to an accessed document, the clusters with these phrases can be determined from the cluster bit vectors for each For each cluster, each phrase that is a member of the cluster is determined by the phrase up in ifis related phrase table that contains the cluster number, or bit vector representation as described above. This cluster number is then added to th ~ 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 ~e used as weights, as described below. Thus, the user model is built from phrases in clusters that are present on a document that the user has expressed an in by accessing the document.
The same general approach can be more precisely focused to capture information where a higher level of interest than merely accessing the document by the user (which the user may do simply to judge if indeed the is relevant). For example, the collection of phrases into the user model may to those documents that the user has printed, saved, stored as a favorite or link, email 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.
[D2ldj When another query is received from the user, the related query phrases Qr a ~ retrieved. These related query phrases Qr are intersected with the phrases listed in th ~ user model to determine which phrases are present in both the query and the user r modgl. A mask bit vector is initialized for the related phrases of ~e query Qr. This bit is a bi-bit vector as described above. For each related phrase Qr of the query that is alsp present in the user model, both of the bits for this related phrase are set in the bit vector. 'The mask bit vector thus represents the related phrases present in both the 4uerv and the user model.
The mask bit vector is then used to mask the related phrase bit vector for each Idocurr~ent in the current set of search results by ANlJing the related phrase bit with the mask bit vector. This has the effect of adjusting the body score and the hit score by the mask bit vector. The documents are then scored far their body and anchor score as before and presented fio the user. This approach essentially requires that a document have the query phrases that are included in the user model in ordel to be highly ranked.
(OZI~~ As an alternative embodiment, which does not impose the foregoing tight l constraint, 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 s gets multiplied by 0 or 1, effectively zeroing or maintaining the counts.
Next, counts themselves are used as weights are also used to multiply the related es for each document that as being scored. This approach has the benefit of ing documents that do not have the query phrases as related phrases to still score Finally, the user model may be limited to a current session, where a on is an interval of lame for active period of time in search, after which session the user model is dumped. Alternatively, the user model for a given user may be persisted over time, and then down-weighted or aged.
l IV. Result Presentation (022~j 'the presentation system 1~0 receives the scored and sorted search results from!the search system 120, and performs further organizational, annotation, and clus~ring operations prior to prig the results to the user. These operations facalil to the user's understanding of the content of the search results, eliminate dup ~cates, and provide a more representative sampling of the search results.
FIG. 7 illus ~ ates the main functional operations of the presentation system 130:
(Q21~] 700: Quster documents aaording to topic clusters lozyj 702: Generate document descriptions j021 704: Eliminate duplicate documents.
j02It3j Each of these operations takes as an input the search results 701 and outputs modified search results 703. As suggested by FIG, 7, the order of these oper ~ Lions is independent, and may be varied as desired for a given embodiment, and J
thus a inputs may be pipelined instead of being in parallel as shown.
1. Dxnamic Taxonomy Generation for Presentation j02I ~ j For a given query, it is typical to return hundreds, perhaps even thousands of documents that satisfy the query. In many cases, certain documents, while haviikg different content from each other, are sufficiently related to form a meaningful groin of related documents, essentially a cluster. Most users however, do not review the first 30 or 40 documents in the search results. Thus, if the first 100 fox example, would come from three clusters, but the next 100 documents an additional four dusters, then without further adjustment, the user will not review these later documents, which in fact may be quite relevant to the query since they represent a variety of different topics related to the query.
Thus, desirable to provide the user with a sample of documents from each cluster, exposing the user to a broader selection of different documents from the search The presentation system I30 does this as follows.
As in other aspects of the system 100, the preseniation system 7.30 makes use o~ the related phrase bit vector for each document d in the search results. More I
spec'I cally, for each query phrase Q and for each document d in Q's posting list, the related phrase bit vector indicates which related phrases Qr are present in the document.
Over~Ithe set of documents in the search results then, for each related phrase Qr, a count is de ~ermined far how many documents contain the related phrase Qr by adding up the bit v ~ ues in the bit position corresponding to Qx. When summed and sorted over the sear results, the most frequently occurring related phrases Qr will be indicated, exh of -ch will be a cluster of documents. The most frequently oowrring related phrase is the first cluster, which takes as its name its related phrase Qr, and so on for the top three to fife clusters. Thus, each of the top clusters has been identified, along with the phrase Qr asi a name or heading far the cluster.
Now, documents from each cluster can be presented to the user in ways. Tn one application, a fixed number of documents from each cluster ran be for example, the 10 top scoring documents in each cluster. 1n another application, a proportional number of docmn.ents from each cluster may be presented.
Thus; if there are 100 documexlts in the search result, with 50 in cluster 1, 30 in cluster 2, 10 in jcluster 3, 7 in cluster 4, and 3 in cluster 5, and its desired to present only 20 I
docu~nts, then the documents would be select as follows: IO documents from cluster l I; 7 documents from cluster 2, 2 documents from cluster 3, and 1 document from cluster I

4. Tt ie documents can then be Shawn to the user, grouped accordingly under the apprppriate duster names as headings.
I
[022~J For example, assume a search query of "blue merle agility training", for which the search system 120 retrieves 100 documents. The search system 120 will have alrea ~ y identified "blue merle" and "agility training" as query phrases. The related phr es of these query phrases as:
j0 J "blue merle":: "Australian Shepherd," "red merle," "tricolor, "aussie";
l j0224,J "agility training":: "weave poles," "teeter," "tunnel," "obstacle,"
"border I
collie,".

The presentation system 130 then determines for each of the above phrases of each query phrase, a count of the number of documents contain such For example, assume that the phrase "weave poles" appears in 75 of the 300 "teeter" appears in 50 documents, "red merle" appears in 50 documents.
first cluster is named "weave poles" and a selected number of documents from that cfluster are presented the second cluster is named "teeter," and selected number are presented as well, and so forth. For a fixed presentation,10 documents from each cluster may be selected. A proportional presentation would use a proportionate nuTnber of document's from each duster, relative to the total number of documextts.
t 2. Topic Based Document Desrxiptions [022C] A second function of the presentation systiem 230 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, amd thus help the user understand what the document is about in a way that is contextually related to the search. The document descriptions can be eithex general or perBOInaT~2ed t0 the user.
a) General Topic Document Descriptions I0227j As before, given a query, the search system 120 has determined the related query phrases Qr and the phrase extensions of the query phrases as well, and I
then identified the relevant documents for the query. The presentation system access each document in the search results and perform the follow operations.
[(1220 First, the presentation system 23i) Tanks the sentences of the document by the n ~ r of instances of query phrases Q related query phrases Qr, and phrase Qp, thereby maintaining for each sentence of a document counts of these Then the sentences are sorted by these counts, with the first sort key the count of query phrases Q, the second sort key being the count of related query Qr, and the final sort key being the count of phrase extensions Qp.
Finally, the top N (e.g., 5) sentences following the sort are used as the of the document. This set of sentences can be formatted and included in the of the document in ii~e modified search results 703. T7zis process is for some number of documents in the search results, and may be done on each time the user requests a next page of the results.
b) Perso alized Topic Based Document Descriptions In embodiments where personalization of the search results is provided, the dbcument descriptions can likewise be personalized to reflect the user interests as in the user model The presentation system 130 does this as follows.
First, the presentation system 13a determines, as before, the related that are relevant tb the user by intersecting the query related phrases Qr with the user model (which lists the phrases occurring in documents accessed by the user).
The presentation system 130 then stable sorts this set of user related Ur according to the value of. the bit vectors themselves, prepextding the sorted list to~'the list of query related phrases Qr, and removes any duplicate phrases. The sort maintains the existing order of equally ranked phrases. This results in a set of phrases which related to the query or the user, called set Qu.
Now, the presentation system 130 uses this ordered list of phrases as the basis for ranking the sentences in each document in the search results, in a manner sito the general document description process described above. Thus, fox a given docmlinent, the presentation system I30 ranks the sentences of the document by the number of instances of each of the user related phrases and the query related phrases Qu, aid sorts the ranked sentences according to the query counts, and finally sorts based: on the number of phrase extensions for each such phrase. Whereas previously the sort keys where in the order of the query phrases Q, related query phrases Qr, and phrasi extension Qp, here the sort keys are in the order of the highest to lowest ranked user related phrases Ur.
I
(~235~ Again, this process is repeated for the documents in the search results (either on demand or aforehand). For each such dorun~ent then the resulting document ties ' tion comprises the N top ranked sentences from the document. Here, these sentez~ces will the ones that have the highest numbers of user related phrases Ur, and I
thus represent the key sentences of the document that express the concepts and topics most l elevant to the user (at least according to the information captured in the user modelj.
3. Duplicate Document Detection and l~iminatiar~~
10236] . In large corpuses such as the Internet, it is quite common for there to be multiple instances of the same document, or portions of a document in many different loca#ic~ns. For example, a given news article produced by a news bureau such as the Associated Press, ntay be replicated in a dozen or more websites of individual newspapers. 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 documents that are likely tn be duplicates or near duplicates of each other, and only include 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 documents that are duplicates of each other. The presentation system 134 provides the functionality as follows.
[023 The presentation system 130 processes each document in the search result set 74i . For each document d, the presentation system ?34 first determines the Iist of related phrases R associated with the document. For each of these related phrases, the presentation system ?34 ranks the sentences of the document according #o the frequency of occ~czrrence of each of these phrases, and then selects the top N (e.g., 5 to ?4) ranking senfiences. This set of sentences is then stored in association with the document. Une way to do this is to concatenate fine selected sentences, and then take use a hash table to store $he document identifier.
Ia238~ Then, the presentation system T30 compares the selected sentences of each document d to the selected sentences of the other documents in the search results 701, and if the selected sentences zonat~ch (within a tolerance}, the documents are presumed to be duplicates, and one of them is removed from the search results.
For example, the presentation system ?30 can hash the concatenated sentences, and if the hash table already has an entry for the hash value, then this indicates that the current document and presently hashed document are duplicates. The presentation system can then update the table with the document 1D of one of the documents.
F~referably, the presentation system 134 keeps the document that has a higher page rank or other queryindependent measure of document significance. In addition" the presentation syste~ri 130 can modify the index 150 to remove the duplicate document, so that it will not appear in future search results for any query.

(0239 The same duplicate elimination process may be applied by the indexing I
systeijn 110 directly. When a document is crawled, the above described document descr;ption process is performed to obtain the selected sentences, and then the hash of these sentences. If the hash table is filled, then again the newly crawled document is deemed to be a duplicate of a previous document. Again, the indexing system 110 can then beep the document with the higher page rank or other query independent measure.
[4240 The present invention has been described in particular detail with respect to one possible embodiment. Those of skill in the art will appreciate that the invention may be practiced in other embodiments. First, the particular naming of the components, capitalization of terms, the attributes, data structures, or any other 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, i the system may be implemented via a combination of hardware and software, as I
described, or entirety in hardware elements. Also, the particular division of functionality between the various system components descn'bed herein is merely exemplary, and not mandatory; functions performed by a single system component may instead be perfozmed by multiple components, and functions performed by multiple components I
may instead performed by a single component.
[0241] Some portions of above description present the features of the present invention in terms of algorithms and symbolic representations of operations on information. These algorithmic descriptions and representations are the means used by those skilled in the dafia processing arts to most effectively convey the substance of their work tao others skilled in the art. These operations,.while described functionally or logically, are understood to be implemented by computer programs. Furthermore, it has t also proven convenient at times, to refer to these arrangements of operations as modules or by functional nannes, without loss of generality.
[0242 Unless specifically stated otherwise as apparent from the above discussion, it is appreciated that throughout the descriptian, discussions utilizing terms such as "processing" or "computing" or "calculating" or "determining" or "displaying"
or they like, refer to the action and processes of a computer system, or similar electronic computing device, that manipulates and transforms data represented as physical (electronic) quantities within the computer system memories or registers or other such infornnation starage, transmission or display devices.
[U243] Certain aspects of the present invention include process steps and instruictions described herein in the form of an algorithm. It should be noted that the process steps and instructions of the present invention mold be embodied in software, firmware or hardware, and when embodied in software, could be downloaded to reside on and be operated from different platforms used by real time network operating systems.
[0244] The present invention also relates to an apparatus for performing the operations herein. This apparatus may be specially constructed for the required porpoises, or it may comprise a general-purpose computer selectively activated or reconfigured by a computer program stored on a computer readable medium that can be accessed by the computer. Such a computer program may be stored in a computer readable storage medium, such as, but is not limited to, any type of disk including floppX disks, 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 sto ' electronic instructions, and each coupled to a computer system bus.
Furth ore, the computers referred to in the specification may include a single I
processor or may be architectures employing multiple processor designs for increased comp~tnng capability.
[0245 The algorithms and operations presented herein are not inherently relateii to any particular computer or other apparatus. Various general-purpose systems may also be used with programs in accordance with the teachings herein, or it may prove;convenient to construct more specialized apparahxs to perform the required method steps. The required structure for a variety of these systems will be apparent to those of skill in the, along with equivalent variations. In addition, the present invention is not described with reference to any particular programming language. It is r appreciated that a variety of programming languages may be used to implement the I
teat 'hiizgs of the present invention as described herein, and any references to speafic languages are provided for disclosure of enablement and best mode of the present invention.
[0246j The present invention is well suited to a wide variety of computer netwrsrk systems over numerous topologies. Within this field, the configuration and management of large networks comprise storage devices and computers that are connnyunicatively coupled to dissimilar computers and storage devices over a network, such as the Internet.
I0247j Finally, it should be noted that the language used in the specification has been principally selected for readability and instructonaI purposes, and may not have been selected to delineate or circumscribe the inventive subject matter.
Accordingly, the disclosure of the present invention is intended to be illustrative, but not limiting, of the scopetof the invention, which is set forth in the following claims.

Claims (14)

1. A method of selecting documents in a document collection in response to a query, the method comprising:
receiving a query;
identifying a plurality of phrases in the query, wherein at least one phrase is a multiple word phrase;
identifying a phrase extension of at least one of the identified phrases; and selecting documents from the document collection containing at one phrase from a set including phrases in the query and the phrase extension.

2. The method of claim 1, wherein selecting documents comprises:
combining a posting list of an identified phrase and a posting list of the phrase extension of the identified phrase to form a combined posting list; and selecting documents appearing in the combined posting list and in the posting lists of the other identified phrases.

3. A method of selecting documents in a document collection in response to a query, the method comprising:
receiving a query;
identifying an incomplete phrase in the query;
replacing the incomplete phrase with a phrase extension; and selecting documents from the document collection containing the phrase extension.

4. The method of claim 3, wherein identifying an incomplete phrase and replacing the incomplete phrase comprise:
identifying a candidate phrase in the query;~
matching the candidate phrase to an incomplete phrase in a list of incomplete phrases; and replacing the candidate phrase with a phrase extension associated with the incomplete phrase.

5. ~The method of claim 3, wherein a phrase extension of an incomplete phrase comprises a super-sequence of the incomplete phrase that begins with the incomplete phrase.

6. ~A method of selecting documents in a document collection in response to a query, the method comprising:
receiving a query including a first phrase and second phrase;
retrieving a posting list of documents containing the first phrase;
for each document in the posting list:
accessing a list indicating related phrases of the first phrase that are present in the document; and responsive to the list of related phrase indicating that the second phrase is present in a document, selecting the document to include in a result to the query, without retrieving a posting list of documents containing the second phrase.

7. The method of claim 6, further comprising:
responsive to the list of related phrases indicating that the second phrase is not present in a document, excluding the document from the result to the query, without retrieving a posting list of documents containing the second phrase.

8. The method of claim 6, further comprising.
responsive to the list of related phrases indicating that the second phrase is not a related phrase of the first phrase, intersecting the posting list of documents for the first phrase and with a posting list of documents for the second phrase to select documents containing both the first phrase and the second phrase.

9. The method of claim 6, further comprising:
storing the list of related phrases for a first phrase with respect to a document in a bit vector, wherein a bit of the bit vector is set for each related phrase of the first phrase that is present in the document, and a bit of the vector is onset for each related phrase of the first phrase that is not present in the document, wherein the bit vector has a numerical value; and scoring a selected document by determining an adjusted value of the bit vector according to the bits set for related phrases of the first phrase that are present in the document.

10. A method of ranking documents included in a search result in response to a query, the query comprising at least one query phrase, the method comprising:
for each document in the search result, accessing a related phrase bit vector for a query phrase, wherein each bit of the bit vector indicates the presence or absence of a related phrase of the query phrase; and sorting the documents in the search results by the value of their related phrase bit vectors, such the document with the highest value related phrase bit vector is ranked highest in the search result.

11. The method of claim 10, wherein:
each bit of the related phrase bit vector is associated with a related phrase of the query phrase; and the bits are ordered such that a most significant bit of the bit vector is associated with a related phrase having a greatest information gain with respect to the query phrase, and a least significant bit is associated with a related phrase having the least information gain with respect to the query phrase.

12. A method of ranking documents included in a search result in response to a query, the query comprising at least one query phrase, the method comprising:
for each document in the search result:
accessing a related phrase bit vector for a phrase of the query, wherein each bit of the bit vector indicates the presence or absence of a related phrase of the query phrase;

for each bit indicating the presence of a related phrase of the query phrase, adding a predetermined number of points associated with the bit to a score for the document; and sorting the documents in the search results by their document scores.

13. The method of claim 22, wherein:
each bit of the related phrase bit vector is associated with a related phrase of the query phrase;
the bits are ordered such that a most significant bit of the bit vector is associated with a related phrase having a greatest information gain with respect to the query phrase, and a least significant bit is associated with a related phrase having the least information gain with respect to the query phrase; and the predetermined number of points associated with each bit range from a most number points associated with the most significant bit to a least number of points associated with a least significant bit.

14. A method of providing an information retrieval system, the method comprising:
automatically identifying valid phrases in a document collection comprising a plurality of documents, wherein the valid phrases contain multiple word phrases;
indexing the documents according to valid phrases contained in the documents;
receiving a search query;
identifying phrases contained in the query;
selecting documents according to the identified phrases; and ranking the selected documents according to the identified phrases.