JP2009238007A - Information retrieval device and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To personalize a retrieval result so that word-of-mouth information is ranked high out of a large number of documents and data. <P>SOLUTION: When retrieval conditions are input, retrieval is performed with documents of close persons as retrieval objects sequentially from persons closer in social relation to a retriever set as a starting point, and scores are computed on the retrieved documents. A distance to the retriever is extended to enlarge a search range. At this time, the maximum score that can be taken by the distance is computed, and when there is no possibility of the maximum score exceeding a minimum score, search is ended, and the N documents ranked high are output as a retrieval result. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to an information search apparatus and program, and more particularly to an information search apparatus and program for searching for information that meets a search condition from a large number of information (documents and data).

  More specifically, the present invention relates to an information search apparatus and program for personalizing search results using other information related to a searcher.

  In particular, the present invention relates to an information search apparatus and program for ranking documents and data using information of a social graph representing a search result.

  As an example of a typical personalized search, if you enter the search condition "Jaguar", it is more appropriate to present animal jaguars to those who like animals, and to people who like cars It is better to present a Jaguar car. However, a search device that does not have a personalization function, such as a Web search engine or a database management system, determines a search result based on an input search condition and accumulated documents / data. .

Further, as an existing search technique related to the personalization function, a method using a bookmark of a user's Web browser (for example, see Non-Patent Documents 1 and 2), a method using a click log (for example, Non-Patent Document 3) Is mentioned.
Glen Jeh, Jennifer Widom, "Scaling personalized web search," International World Wide Web conference, Pages: 271-279,2003 "Personalized search." Communications of the ACM vol. 45, No. 9, Pages: 50-55, 2002. Zhicheng Dou, Ruihua Song, Ji-Rong Wen, "A Large-scale Evaluation and Analysis of Personalized Search Strategies", International World Wide Web Conference, Pages: 581-590 2007.

  However, in the search technology that does not have the above personalization function, there is a problem that different search interests cannot be reflected in the search result for each search with respect to the same search condition. In addition, as for a technology having a personalization function, a search technology for personalizing a search result using a distance between a person (copyright information etc.) related to a document or data and a searcher has not been known in the past.

  One way to get information that people can trust is through word-of-mouth communication that propagates information from friends to friends.

  The present invention has been made in view of the above points, and an object of the present invention is to provide an information search apparatus and program capable of personalizing search results so that word-of-mouth information is ranked higher.

  FIG. 1 is a diagram for explaining the principle of the present invention.

The present invention is an information retrieval method for retrieving a text document group,
Social graph storage means for storing a social graph indicating a relationship between people;
Search target document storage means storing the search target document;
In an information retrieval apparatus having a person / document relation information storage unit storing relation information between a document and a person,
When a searcher ID for uniquely identifying a searcher and a search condition are input (step 1),
The social graph storage means is searched based on the searcher ID, the ID of the person who is close to the searcher of the searcher ID is searched (step 2), and the person / document relationship is based on the ID of the person who is close to the distance Searching for a document related to the ID of the person who is close to the distance from the information storage means (step 3), searching the search target document storage means with search conditions using the document as a search target document (step 4);
An output step (step 5) for outputting the top N searched documents to the information terminal of the searcher is performed.

  FIG. 2 is a principle configuration diagram of the present invention.

The present invention (Claim 1) is an information retrieval apparatus for retrieving a text document group,
Social graph storage means 3 storing a social graph indicating a relationship between people,
Search target document storage means for storing the search target document, and 5,
A person / document relation information storage means 4 storing relation information between a document and a person;
When a searcher ID for uniquely identifying a searcher and a search condition are input, the social graph storage unit 3 is searched based on the searcher ID, and an ID of a person who is close to the searcher of the searcher ID is selected. A search is performed to search for a document related to the ID of the person who is near the distance from the person / document relation information storage unit 4 based on the ID of the person who is close to the distance. Search means 1 for searching the document storage means 5;
Output means 13 for outputting the top N searched documents to the searcher's information terminal.

Further, the present invention (Claim 2) is provided in the search means 1,
Score calculating means for calculating the score value of the document using the value of the degree of matching between the document and the search condition, and the value of the distance from the searcher to the person related to the document in the social graph of the social graph storage means 3 In addition,
The distance between the searcher and the person involved in the document is
It is defined by the number of hops (internode distance) in the graph structure, or the linear sum of additional factors of reference and information.

In addition, the present invention (Claim 3) further includes an index construction unit that constructs an index for a document group related to the person, and stores the index group in the index storage unit before the search.
Search means 1
In the social graph stored in the test social graph storage means, the social graph is searched centering on the searcher, the index of the searched person is obtained by referring to the index storage means, and the input search condition and the index Means for searching the search target document storage means using.

Further, the present invention (Claim 4) is provided in the search means 1,
When the number of searches is specified,
When searching the social graph of the social graph storage means 3 starting from the searcher with an expanded range,
Comparing the document with the lowest score in the middle of the search calculated by the score calculation means with the maximum score that can be taken by the document related to the person to be searched for, it was determined that the minimum score did not exceed the maximum score Means for terminating the search of the social graph at the time.

  The present invention (Claim 5) is an information search program for causing a computer to function as each means constituting the information search apparatus according to any one of Claims 1 to 4.

  Conventionally, a bookmark of a Web browser or a click log of a user is used. In the present invention, ranking is performed using a relationship between people, so that a person who is close to a searcher can be ranked. It becomes possible to rank related documents at the top of the search results.

  Further, in the present invention, in addition to the value of the degree of matching between the document used conventionally and the search condition, the score value is calculated based on the distance between the searcher in the social graph and the person related to the document, and the search result Determine the document order. Thereby, the ranking of search results can be performed in consideration of the distance that information propagates to people.

  Further, in the present invention, an index is constructed in advance for a group of documents related to the person in units of people, and at the time of search, the searcher is set as the starting point. Use the index to search for documents with high relevance, sequentially search the social graph from the searcher to a person who is far away, search the document using the index of the search target person, and select search result candidates To construct. Thereby, a search process can be performed at high speed by using the existing index technology.

  Furthermore, in the present invention, when a searcher performs a search by specifying the number of search cases, when searching the social graph far away from the searcher, a search is made for a document with the lowest score obtained in the middle of the search The maximum score that a document related to a person can take is compared, and when it is determined that the latter score does not exceed the former score, the search is terminated. Thereby, it is possible to further speed up the search process by avoiding the entire search of information.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

  FIG. 3 is a configuration diagram of a social information search apparatus according to an embodiment of the present invention.

  The social information search device 10 includes an information search unit 1, an index construction unit 2, a social graph storage unit 3, a person-document relationship information storage unit (hereinafter simply referred to as a relationship information storage unit) 4, and a search target document storage unit 5 And an index storage unit 6.

  Here, the social graph storage unit 3, the relationship information storage unit 4, the search target document storage unit 5, and the index storage unit 6 are storage means such as a disk device. The index storage unit 6 has a data structure for speeding up the search process for transposed files, B + -trees, and the like.

  FIG. 4 shows an example of a social graph in the social graph storage unit in one embodiment of the present invention.

  In the figure, the dots represent people, and the relationship from person to person is represented as an arrow connecting the points. For simplification, a bidirectional arrow is expressed using a line having arrows at both ends, and the person's ID and the distance from person to person are stored in the social graph storage unit 3.

  For example, in social networking services such as mixi, one person A records that he saw another person's B diary as a “footstep”, which can be expressed as a relationship from A to B. it can. Similarly, when A adds “comment” to the diary of another person B, it can be regarded as a relationship from A to B. When A and B are in a “friend” relationship, it can be expressed as a bidirectional relationship between A and B. When constructing a social graph, it may be constructed using any one or more of these “footstep”, “comment”, and “friendship” information. The distance from person to person is determined by the number of hops (distance between nodes) in the social graph, or a distance defined by the linear sum of factors of reference (footstep) and information addition (comment) There are either methods defined by:

  For example, the distance from A to B relating to the search condition Q is defined by the following formula when A and B are adjacent in the social graph.

Formula 1)
Distance (A, B, Q) =
α × {number of documents of all friends of A that match Q to which A added information}
/ {Number of B documents that match Q to which A added information}
+ Β × {number of documents of all friends of A that match Q referenced by A}
/ {Number of B documents matching Q referenced by A}
Here, α and β are numerical values for linear summation and normalization.

  If A and B are not adjacent to each other in the social graph, the reachable path from A to B is obtained on the social graph, and the distance between any two nodes constituting the path is summed up by the formula 1. Is defined as the distance from A to B. When there are a plurality of reaching paths, the sparse minimum value is defined as the distance from A to B.

  As another expression of distance, the distance defined by the number of hops in the graph is given by the following equation.

Formula 2)
Distance (A, B) = {shortest path length from A to B}
The feature that the document score should satisfy must satisfy that the document score is higher if the document is more suitable for the search condition, and the document score is lower as the distance from the searcher to the author of the document is larger. There is. One example is a document score defined by the following equation.

Formula 3)
Document score (t) = (Fitness of document t to search condition) / (Distance from searcher to author of document t)
Note that “the fitness of the document t from the searcher” in Equation 3 is the fitness calculated by an existing method such as a transposed file, and the “distance from the searcher to the author of the document t” is expressed by a social graph. This is the numerical value obtained.

  FIG. 5 shows an example of a person-document relationship information storage unit in an embodiment of the present invention. The person-document relationship information storage unit 4 stores the person ID of the person, the relationship (distance) between the person and the document ID related to the person ID.

  FIG. 65 shows an example of an index storage unit according to an embodiment of the present invention, which is an example of an index for a document related to a certain person A. In the figure, document IDs (d1, d5,...) Related to a certain word, and the values (t15, t8) relating to the document such as the appearance frequency of words in the document are stored.

  FIG. 7 shows the configuration of the information search unit in one embodiment of the present invention.

  The information search unit 1 shown in the figure includes a search unit 11, a score calculation unit 12, a search result selection unit 13, a memory A14, and a memory B15.

  When a searcher 11 receives information (hereinafter referred to as a searcher ID) and a search condition from the information terminal 20, the searcher 11 stores a social graph, a person and a document stored in the social graph storage unit 3. The relationship order information between the person and the document stored in the relationship information storage unit 4 and the index information stored in the index storage unit 6 are retrieved and stored in the memory A14.

  The score calculation unit 12 calculates a score using the distance between the searcher and the searched person, and stores it in the memory B15 as a search result candidate.

  The search result selection unit 13 selects the top n search candidates in the memory B15 when the highest possible score of the index in the index storage unit 6 does not exceed the top n scores calculated by the score calculation unit 12. And output to the information terminal 20.

  The index construction unit 2 reads a document from the search target document storage unit 5 before the search process in order to speed up the search process of the information search unit 1, and for a document group related to that person in units of people. An index is constructed and stored in the index storage unit 6. As a result, the information search unit 1 searches for a document having a high degree of fitness using the searcher as a starting point and using the input search condition and the index of a person adjacent to the searcher in the social graph.

  Next, the operation in the above configuration will be described.

  FIG. 8 is a flowchart of the index construction unit in one embodiment of the present invention.

  Step 101) The index construction unit 2 reads the search target document from the search target document storage unit 5.

  Step 102) Next, the relationship information between the person and the document is read from the relationship information storage unit 4 between the person and the document, and the documents are grouped for each person concerned.

  Step 103) A person who has not completed the construction of the index is selected, and an index is constructed for the group of documents related to that person and stored in the index storage unit 6. As an index building method, there are a transposed file method, a B + -tree method, and the like.

  Step 104) It is determined whether or not the processing has been completed for all search gymnastic documents. If it has been completed, the index construction processing is terminated, and if it has not been completed, the processing proceeds to Step 105.

  Step 105) A person who has not finished processing is selected from the relation information between the person and the document, and the process proceeds to Step 103.

  Next, the processing of the information search unit 1 will be described assuming that the above index construction processing has been completed.

  FIG. 9 is a flowchart of the information search unit according to the embodiment of the present invention.

  Step 201) Information (searcher ID) for specifying a searcher, a search condition (search formula) input by the searcher, and the number of output higher ranks (N) are input from the information terminal 20.

  Step 202) The search unit 11 of the information search unit 1 acquires the social graph from the social graph storage unit 3 based on the searcher ID, sets the starting point of the person search to the variable S, and selects the search result candidate. Initialize to empty (= {}). S is a variable that holds a set of searched people, and is stored in the memory A14 of the information search unit 1.

  Step 203) The search unit 11 identifies a person s (but not included in S) that is closest to any person included in the variable S from the social graph and adds the person s to the variable S.

  Step 204) Based on the search condition input as s, the index storage unit 6 is searched, and the score calculation unit 12 calculates the score for the obtained document using the above formula 3), and performs the search. The result is stored in the memory B15. As a method of processing the search condition using the index, there are a transposed file method and a B + -tree method.

  Step 205) Comparing the minimum score of the search result candidates so far with the maximum score of the document obtained by searching in the future. If the latter score does not exceed the former score (Yes), the search is terminated. When that is not right (No), the search of the social graph memory | storage part 3 is continued further.

  The maximum score of the document obtained by searching in the future can be derived from Equation 3 using the distance obtained from Equation 1 or Equation 2 described above and the upper limit value of the fitness that the document can take.

  Step 206) Finally, the search result selection unit 13 outputs the search result candidate stored in the memory B15 to the information terminal 20 to present the result to the searcher.

  Next, the information search process will be described in detail.

  In the following, the number of hops of the social graph is used as the distance of the person, the searcher is A, the search condition indicates a condition for searching for a document including the word “business trip”, and the search result is for the top 20 items. explain.

  FIG. 9 is a diagram (step 1) showing the detailed operation of the search process according to the embodiment of the present invention.

  When the information (searcher ID: A) for specifying the searcher and the search conditions input by the searcher A are input from the information terminal 20, the search unit 11 of the information search unit 1 receives the searcher ID (A). Based on the social graph storage unit 3, a social graph with the number of hops = 1 is acquired. In the example of FIG. 9, G, H, C, and E are acquired. Next, the search unit 11 refers to the index storage unit 6 based on the acquired social graph (G, H, C, E), and the number of hops = 1 from A with the searcher A as the center. A search result candidate is acquired based on the index relating to the documents of the persons G, H, C, and E and the input search condition “business trip”, and stored in the memory A14.

  Next, the score calculation unit 12 reads a search candidate from the memory A14 and calculates a document score using “1” which is a distance from A. The search result candidates are sorted based on the document score and stored in the memory B15.

  FIG. 10 is a diagram (step 2) showing the detailed operation of the search process in one embodiment of the present invention.

  The search unit 11 searches the social graph storage unit 3 with the number of hops = 2 from the searcher A, searches the index storage unit 6 based on the acquired people J, K, D, B, and F, and the obtained index Is searched based on the search condition “business trip” and stored in the memory A14.

  The score calculation unit 12 acquires search candidates from the memory A14, calculates the score of the document using the distance “2” from A, sorts the search candidates based on the score, and selects the search candidates together with the scores. It adds to the search result candidate of memory B15. It is assumed that the 20th-ranked document score, which is the lowest score of search result candidates at this point, is 3.4.

  FIG. 11 is a diagram (step 3) showing the detailed operation of the search process according to the embodiment of the present invention.

Here, in the same manner as described above, the search is performed based on the search condition “business trip” using the index of the person who is searched from the searcher A with the number of hops = 3. Assuming that the upper limit value of the fitness that the retrieved document can have is 1.0, the number of hops here is “3”.
Document score = 1.0 / 3 = 3.3.

  If the current 20th-ranked document score is 3.4, it can be determined that no document exceeding the score of 3.4 is found after the number of hops of 3. In this case, the search results up to the 20th place are confirmed, the search of the social graph is terminated, and the search result candidates up to the process of FIG. 12 are returned to the information terminal 20 as the final results.

  The operations of the components shown in FIGS. 3 and 7 can be constructed as a program and installed in a computer used as the information search apparatus to be executed or distributed via a network.

  Further, the constructed program can be stored in a portable storage medium such as a hard disk, a flexible disk, or a CD-ROM, and can be installed or distributed in a computer.

  The present invention is not limited to the above-described embodiment, and various modifications and applications can be made within the scope of the claims.

  The present invention is applicable to a document retrieval technique having a relationship between a person and a document.

It is a figure for demonstrating the principle of this invention. It is a principle block diagram of this invention. It is a lineblock diagram of a social information search device in one embodiment of the present invention. It is a figure which shows the structure of the social graph in one embodiment of this invention. It is an example of the relationship information storage part of the person and document in one embodiment of this invention. It is an example of the index memory | storage part in one embodiment of this invention. It is a block diagram of the information search part in one embodiment of this invention. It is a flowchart of the index construction part in one embodiment of this invention. It is a flowchart of the information search part in one embodiment of this invention. It is a figure (step 1) which shows the detailed operation | movement of the search process in one embodiment of this invention. It is a figure (step 2) which shows the detailed operation | movement of the search process in one embodiment of this invention. It is a figure (step 3) which shows the detailed operation | movement of the search process in one embodiment of this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Search means, Information search part 2 Index construction part 3 Social graph memory | storage means, Social graph memory | storage part 4 Person / document relation information storage means, Person and document relation information storage part 5 Search object document storage means, Search object document storage part 6 Index storage unit 11 Search unit 12 Score calculation unit 13 Output means, search result selection unit 14 Memory A
15 Memory B

Claims (5)

An information retrieval device for retrieving a text document group,
Social graph storage means for storing a social graph indicating a relationship between people;
Search target document storage means storing the search target document;
A person / document relation information storage means for storing document / person relation information;
When a searcher ID for uniquely identifying a searcher and a search condition are input, the social graph storage unit is searched based on the searcher ID, and an ID of a person who is close to the searcher of the searcher ID is selected. Search, search for a document related to the ID of the person who is near the distance from the person / document relation information storage means based on the ID of the person who is close to the distance, Search means for searching the search target document storage means;
An output means for outputting the top N searched documents to the information terminal of the searcher;
An information retrieval apparatus comprising: The search means includes
Score calculation means for calculating the score value of the document using the value of the degree of matching between the document and the search condition and the value of the distance from the searcher to the person related to the document in the social graph of the social graph storage means Further comprising
The value of the distance of the person related to the document from the searcher is:
The information search apparatus according to claim 1, wherein the information search apparatus is defined by a hop count (internode distance) in a graph structure or a linear sum of additional factors of reference and information. Prior to the search, an index construction unit for constructing an index for a document group related to the person in units of people and storing the index group in the index storage unit,
The search means includes
In the social graph stored in the social graph storage means, the social graph is searched with the searcher as a center, the index of the searched person is obtained with reference to the index storage means, and the input search condition 2. An information search apparatus according to claim 1, further comprising means for searching said search target document storage means using said index. The search means includes
When the number of searches is specified,
When searching the social graph of the social graph storage means starting from the searcher with an expanded range,
The lowest score document in the middle of the search calculated by the score calculation means is compared with the maximum score that can be taken by the document related to the person to be searched, and it can be determined that the minimum score does not exceed the maximum score. The information search device according to claim 1, further comprising means for terminating the search of the social graph at a point in time.   An information search program for causing a computer to function as each means constituting the information search device according to any one of claims 1 to 4.

Images