JP2013156876A - Recommendation query extraction device, method, and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To enable a replacement type query recommendation, even though a clock log has a small storage capacity, and preferentially display candidates having a high probability to be selected.SOLUTION: The recommendation query extraction device according to the present invention acquires a query input from a user and generates a bipartite graph from a URL and a query obtained by retrieving click log storage menas (queries and URLs). At that time, when the input query is not present in the click log storage means, the recommendation query extraction device acquires a URL corresponding to the query via a search engine and adds it to the bipartite graph along with a lower limit (n-1) number of queries included in the bipartite graph. The recommendation query extraction device calculates respective scores of the queries included in the bipartite graph, selects a predetermined number of queries in order of upper scores, and outputs them.

Description

  The present invention relates to a recommended query extracting apparatus, method, and program, and more particularly, to a recommended query extracting apparatus, method, and program for presenting a query that is highly likely to obtain a search result intended by a user.

  The query entered by the user may not be appropriate for obtaining the search result intended by the user, but by selecting a recommended query, the user can easily change the entered query and restart the search. Can try. For example, it is possible to recommend a query of “coffee bean type” or “bean brand” to a user who has entered a query of “bean type” in order to search for documents related to [type of coffee beans]. .

  Hereinafter, a conventional technique for recommending a query will be described. First, terms used in this specification will be described.

  A “query” refers to a character string that a user enters into the system to search for a document. A document to be searched is a document including a keyword included in a query, and when the query includes a plurality of keywords, it indicates a document including the keywords simultaneously. For example, if a query “coffee bean type” is entered in order to search for documents related to coffee bean types, documents including both the keywords “coffee bean” and “type” can be searched.

"Click log" refers to a query that is stored in a storage means such as a hard disk, entered by the user in the past, and a URL selected by the user from the search results output by the system for that query. Point to. For example, when a user A inputs a query a to select URL 1 and another user B inputs a query b to select URL 2,
{Query a, URL1}
{Query b, URL2}
Is retained.

  “Score” refers to an index indicating the possibility that the user will select a query when a query is recommended to the user. For example, if the query “bean knowledge” has a score of 0.9 and the query “coffee bean type” has a score of 1.5 for the input query “bean type”, the latter is calculated by taking the score into account. Priority can be recommended.

  As a first conventional technique for recommending a query, a click log consisting of a query entered by a plurality of past users and a URL selected by the user from the search results output by the system for that query There is a technique for constructing a bipartite graph when an input query is included and generating a query as a recommendation candidate (see, for example, Non-Patent Document 1). Here, the bipartite graph is a graph that can divide a vertex set into two subsets so that there are no edges between vertices in each set in graph theory. FIG. 1 shows the configuration of the query extraction apparatus of the technology. The apparatus shown in FIG. 1 includes a query input unit 1 that acquires a query input by a user, a bipartite graph construction unit 2 that constructs a bipartite graph using the input query, and a click log storage in which a click log is stored. Unit 5, parameter storage unit 4 that holds parameters used by each component in the apparatus, score calculation unit 7 that calculates a score for each query included in the bipartite graph, parameters of the parameter storage unit 4 based on the scores The recommended query output unit 8 extracts the number of queries specified in.

  As shown in FIG. 2, the parameter storage unit 4 holds the following contents as parameters.

n: Lower limit on the number of queries to be included in the bipartite graph;
m: number of iterations in score calculation;
k: number of recommended queries;
The operation of the bipartite graph construction unit 2 is shown in FIG.

  First, the bipartite graph construction unit 2 receives the query q from the query input unit 1 and creates a bipartite graph G including only q (step 1). Next, it is determined whether or not q exists in the click log storage unit 5 (step 2), and if it exists (step 3, Yes), n− is determined based on the access relation between the query and the URL, starting from q. One query and the URL accessed from the query are added to the bipartite graph G (step 4).

  Further, as a second conventional technique, there is a technique for recommending candidates by using a URL searched by a search engine even if an input query is not included in the click log (see, for example, Non-Patent Document 2). .

  Further, as a third conventional technique, there is a technique for estimating the similarity between queries using the similarity of search results for each query and recommending a query similar to an input query to a user (for example, non-search (See Patent Document 3).

Qiaozhu Mei, Dengyong Zhou, Kenneth Church: Query suggestion using hitting time, In Proceedings of CIKM '08, pages 469-478, ACM, 2008. US 2005/0055341 A1: System and method for providing search query refinements, Paul Haahr et al. Japanese Patent Laid-Open No. 2001-202390

  However, the first conventional technique described above can perform a replacement type query recommendation that does not depend on the performance of the search engine for the query included in the click log, but in order to generate a recommended candidate, the click log There is a problem that a recommendation log cannot be generated unless an input query is included in the click log, and a click log having a sufficient capacity is required.

  The second conventional technique can recommend a query for a query that is not included in the click log. However, since the URL searched by the search engine is required, the performance of the search engine is a result of the query recommendation. Cannot recommend candidates that have a large impact and are likely to be selected.

  In addition, the third conventional technique described above can perform query recommendation based on the similarity of search results to a query, but does not use a result determined by a human (corresponding to a click log in the present invention). May greatly affect the result of query recommendation and may not recommend candidates that are likely to be selected.

  The present invention has been made in view of the above points, and allows replacement type query recommendation even when the storage capacity of the click log is small, and can prioritize candidates that are likely to be selected. It is an object of the present invention to provide a recommended query extraction device, method, and program.

In order to solve the above problems, the present invention (Claim 1) provides a recommended query for presenting a query that is highly likely to obtain a search result intended by the user from the click log based on the query input by the user. An extraction device,
Query input means for obtaining a query input by a user;
A click log storage means for storing a click log comprising a set of a query inputted in the past and a URL accessed by the query;
A bipartite graph construction means for generating a bipartite graph from the URL obtained by searching the click log storage means based on the inputted query, and the query;
Score calculating means for calculating respective scores for the queries included in the bipartite graph;
A recommended query output means for selecting and outputting a predetermined number of queries in order from the top of the score;
Have
The bipartite graph construction means includes:
When the input query does not exist in the click log storage unit, the URL includes a URL search unit that acquires a URL corresponding to the query via a search engine and adds the URL to the bipartite graph. .

The present invention (Claim 2) provides the URL search means,
The URL acquired through the search engine is stored in the storage means, and the lower limit value n−1 queries to be included in a predetermined bipartite graph starting from the first URL, and the n−1 Means for adding to the bipartite graph the URL accessed from the query.

The present invention (Claim 3) provides the bipartite graph construction means,
An evaluation value of the bipartite graph obtained by dividing the number of queries included in the bipartite graph by a predetermined number k of recommended queries is obtained, and a bipartite graph in which the evaluation value is larger than a lower limit value of the predetermined evaluation value is obtained. Means for outputting.

  According to the present invention, for an input query that is not included in the click log, a replacement-type query recommendation that recommends a query in which a part or all of the query is replaced is possible, so the storage capacity of the click log is small. However, replacement query recommendation is possible.

  In addition, when the input query is included in the click log, the score of the query to recommend can be calculated using only the result determined by humans without depending on the performance of the search engine. Can be displayed first.

1 is a configuration diagram of a first conventional query extraction device. FIG. FIG. 5 is a diagram showing parameters in a parameter storage unit of the first conventional technique. It is a flowchart of the bipartite graph construction part of the 1st prior art. It is a block diagram of the query extraction apparatus in one embodiment of this invention. It is an example of the parameter memory | storage part in one embodiment of this invention. It is a storage example of the click log memory | storage part in one embodiment of this invention. It is a flowchart of the bipartite graph construction part in one embodiment of this invention.

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

  FIG. 4 shows the configuration of the query extraction apparatus in an embodiment of the present invention.

  The query extraction device shown in FIG. 1 includes a query input unit 1, a bipartite graph construction unit 20, a search engine 30, a parameter storage unit 40, a click log storage unit 5, an index storage unit 60, a score calculation unit 7, and a recommended query output unit. 8 is composed.

  The query input unit 1, the click log storage unit 5, the score calculation unit 7, and the recommended query output unit 8 are the same as those in the first conventional technique.

  The query input unit 1 receives a query input by a user.

  The bipartite graph construction unit 20 receives the input query, generates a bipartite graph from the URL obtained by searching the click log storage unit 5 based on the query and the query, and the click log storage unit 5 If there is no query input to the URL, a URL is obtained using the search engine 30 and a bipartite graph is constructed. Details of the processing will be described later.

  As shown in FIG. 5, the parameter storage unit 40 includes a lower limit n of the number of queries to be included in the bipartite graph, a lower limit u of the evaluation value of the bipartite graph, the number m of iterations in the score calculation, and the number of recommended queries. Store k. These parameters are determined in advance according to the computer resources and the purpose of use at the time of mounting, and are stored in the parameter storage unit 40 before the query is input to the query input unit 1.

  The index storage unit 60 holds heading information of many documents to be searched and is referred to by the search engine 30 in order to efficiently collate the query with the document in the document search. It is assumed that the creation and update of the heading information is performed regardless of the query recommendation and is constructed in advance. The creation and update processing of the header information can be realized, for example, by the method described in Document 1 “B. Brin, L. Page: The anatomy of a large-scale hypertextual Web search engine, 1998”. is there. Note that the creation / update of heading information is outside the scope of the present invention, and thus detailed description thereof is omitted.

  The click log storage unit 5 holds a query input by a plurality of past users to search for a document and a URL selected by the user from the search results obtained for the query. For example, when a user A in the past inputs a query a and selects URL1, and then user B inputs a query b and selects URL2, the click log storage unit 5 stores the query as shown in FIG. A pair of a and URL1 and a query b and URL2 are held.

  The search engine 30 uses the index storage unit 60 to search for documents including keywords included in the query q, and outputs a list of URLs of those documents as a search result. For example, it can be realized by the method shown in FIG.

  The score calculation unit 7 calculates each score for queries other than the query q included in the bipartite graph g. In the present invention, the bipartite graph output from the bipartite graph construction unit 20 has an evaluation value equal to or greater than a predetermined value. Specifically, it can be realized by using the technique of document 2 “Qiaozhu Mei, Dengyong Zhou, Kenneth Church: Query suggestion using hitting time, In Proceedings of CIKM '08, pages 469-478, ACM, 2008.”.

  The recommended query output unit 8 selects k queries (the lower limit value of the number of queries to be included in the bipartite graph) based on the score obtained by the score calculation unit 7 and outputs them as recommended queries. Specifically, the set Q of the queries to which scores are assigned is rearranged in the order of the scores, k queries are selected in descending order of possibility of being selected by the user, and are output as recommended queries. Specifically, it can be realized by using the technique of Document 2.

  Next, the processing of the bipartite graph construction unit 20 will be described in detail.

  FIG. 7 is a flowchart of the bipartite graph construction unit in one embodiment of the present invention.

  Step 100) The bipartite graph construction unit 20 acquires the query q from the query input unit 1, and creates a bipartite graph G including only q.

  Step 110) Referring to the click log storage unit 5 based on the query q, it is determined whether or not the query q exists.

  Step 120) If the query q exists, the process proceeds to step 130, and if not, the process proceeds to step 140.

  Step 130) Starting from the query q, based on the access relation between the query and the URL, add n−1 queries and the URL accessed from the query to the bipartite graph G on the memory (not shown), The process ends.

  Step 140) If the query q is not stored in the click log storage unit 5, the URL corresponding to the query q is received via the search engine 30, and the URL counter is set to i = 1. The acquired URL is stored in a memory (not shown).

  Step 150) Starting from the i-th URL, based on the access relation between the query and the URL, the i-th URL of the search result by the search engine 30 is used as the starting point of the access relation, and the n-1 queries and the query are accessed. The clicked URL is extracted from the click log storage unit 5 and added to the bipartite graph G. Note that n is the lower limit of the number of queries to be included in the bipartite graph stored in the parameter storage unit 40.

Step 160) The evaluation value of the bipartite graph G is calculated. Evaluation value is
Evaluation value = | Q | / k
However, | Q | is the number of queries included in the bipartite graph G, and k is the number of recommended queries specified by the parameters stored in the parameter storage unit 40.

  Step 170) It is determined whether or not the evaluation value obtained in Step 160 is equal to or higher than the lower limit u of the evaluation value of the graph G stored in the parameter storage unit 40. If the value is smaller, the process proceeds to step 180.

  Step 180) Set the counter to i = i + 1 and return to Step 150.

  In the present invention, even if the input query does not exist in the click log storage unit 5, a bipartite graph can be obtained by acquiring a list of URLs related to the input query through the search engine 30 in the above steps 110 to 130. Can be built. Since the click log is generated when a human selects a URL from the search result for the input query, the URL obtained in step 140 can be considered as a click log candidate for the input query. can do. As a result, the bipartite graph can be constructed even when the click log capacity that can be searched by the query extraction device is small (for example, only during the previous month) and there is a high possibility that the input query does not exist in the click log. If a bipartite graph can be constructed, query recommendation can be performed by performing the processing of the score calculation unit 7 in the same manner as in the prior art.

  On the other hand, when a query exists in the click log storage unit 5, a graph is constructed based only on the click log, so that a query that can reflect human judgment and is highly likely to be selected is recommended. be able to.

  Furthermore, by calculating an evaluation value and outputting a bipartite graph of a predetermined value or more to the score calculation unit 7, it is possible to prioritize and output candidates that are likely to be selected.

  The operations of the query input unit 1, the bipartite graph construction unit 20, the score calculation unit 7, and the recommended query output unit 8 shown in FIG. 4 are constructed as a program, and installed and executed on a computer used as a query extraction device. Or can be distributed via a network.

  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.

DESCRIPTION OF SYMBOLS 1 Query input part 2,20 2 parts Graph construction part 4,40 Parameter storage part 5 Click log storage part 7 Score calculation part 8 Recommended query output part 30 Search engine 60 Index storage part

Claims (7)

A recommended query extraction device for presenting a query that is highly likely to obtain a search result intended by a user from a click log based on a query input by the user,
Query input means for obtaining a query input by a user;
A click log storage means for storing a click log comprising a set of a query inputted in the past and a URL accessed by the query;
URL obtained by searching the click log storage means based on the input query, bipartite graph construction means for generating a bipartite graph from the query, and respective scores for the queries included in the bipartite graph A score calculating means for calculating
A recommended query output means for selecting and outputting a predetermined number of queries in order from the top of the score;
Have
The bipartite graph construction means includes:
When the input query does not exist in the click log storage unit, the URL includes a URL search unit that acquires a URL corresponding to the query via a search engine and adds the URL to the bipartite graph. Recommended query extraction device. The URL search means
The URL acquired through the search engine is stored in the storage means, and the lower limit value n−1 queries and the n−1 number of queries to be included in a predetermined bipartite graph starting from the first URL The recommended query extraction device according to claim 1, further comprising means for adding a URL accessed from a query to the bipartite graph. The bipartite graph construction means includes:
An evaluation value of the bipartite graph obtained by dividing the number of queries included in the bipartite graph by a predetermined number k of recommended queries is obtained, and a bipartite graph in which the evaluation value is larger than a lower limit value of the predetermined evaluation value is obtained. The recommended query extraction device according to claim 1, further comprising means for outputting. A recommended query extraction method for presenting a query that is highly likely to obtain a search result intended by a user from a click log based on a query input by the user,
Query input means for obtaining a query input by a user;
A click log storage means for storing a click log comprising a set of a query inputted in the past and a URL accessed by the query;
A bipartite graph construction means for generating a bipartite graph from the URL obtained by searching the click log storage means based on the inputted query, and the query;
Score calculating means for calculating respective scores for the queries included in the bipartite graph;
A recommended query output means for selecting and outputting a predetermined number of queries in order from the top of the score;
In a device having
If the input query does not exist in the click log storage unit, the bipartite graph construction unit obtains a URL corresponding to the query via a search engine and adds it to the bipartite graph A recommended query extraction method comprising performing steps. In the URL search step,
The URL acquired through the search engine is stored in the storage means, and the lower limit value n−1 queries and the n−1 number of queries to be included in a predetermined bipartite graph starting from the first URL The recommended query extraction method according to claim 4, wherein a URL accessed from a query is added to the bipartite graph. The bipartite graph construction means obtains an evaluation value of the bipartite graph obtained by dividing the number of queries included in the bipartite graph by a predetermined number k of recommended queries, and the evaluation value is a lower limit of the predetermined evaluation value. The recommended query extraction method according to claim 4, further comprising a step of outputting a bipartite graph larger than the value. Computer
A recommended query extraction program for functioning as each means of the recommended query extraction device according to claim 1.

Images