JP5519406B2 - Server apparatus, genre score calculation method, and program - Google Patents

Description

  The present invention relates to a technique for estimating a genre of a search query.

  Search services in the network are used. In order to obtain the desired information more efficiently, the user of the search service needs to create a more appropriate query by combining a plurality of words. In order to improve user convenience, a technique for providing a recommendation query to an input query is known (for example, Patent Documents 1 and 2).

  Patent Document 1 discloses an information search server. In Patent Literature 1, the information search server has a database including operation history information (user ID, investigation target, operation content, operation frequency, evaluation value) by a user. The information search server extracts a web page group for the input request information (product name or genre name to be searched), classifies the extracted web page group into a category, and transmits it to the user (FIG. 7). Patent Document 2 discloses a technique for calculating a score related to a search query. Specifically, Patent Document 2 discloses that a plurality of queries are linked through a URL, a score is calculated for a combination of queries, and another query is recommended from one query based on the score. (Paragraph 0077, FIG. 11).

JP 2004-326537 A JP 2009-252070 A

Even with the techniques described in Patent Documents 1 and 2, the genre score cannot be calculated using the genre scores of other queries.
In contrast, the present invention provides a technique for calculating a genre score using a genre score of another query.

The present invention provides first acquisition means for acquiring a graph in which a plurality of nodes including a query node indicating a query used for a search and a relay node indicating other items related to the search other than the query are connected by an edge, and a query The genre score and the query corresponding to the specified genre identifier are acquired from the genre dictionary including the genre identifier indicating the genre of the query and the genre score indicating the degree of relevance between the query and the genre. 2 acquisition means and a node identified from target nodes indicating queries whose genre scores are not described in the genre dictionary among query nodes included in the graph acquired by the first acquisition means as a base node; Other query nodes connected at the edge via the base node and the relay node The genre score of the query is used from the genre scores acquired by the second acquisition means, and the genre score of the query of the base node is calculated. The base node is updated until the end condition is satisfied. A server device is provided that includes calculation means for repeatedly performing genre scores of a plurality of queries.
According to this server device, a genre score can be calculated using a genre score of another query.

In a preferred aspect, the plurality of nodes are a first location node indicating first location information indicating a location of a document browsed by the user as a result of the search using a hierarchical structure, and a higher rank than the first location information. 2 including a second location node indicating location information and a user node indicating the name of the searched user, wherein the relay node is the first location node, and the processing is connected to the relay node at an edge The genre score of the base query may be calculated using the genre score of a query indicated by the second location node or another query node connected to the user node by an edge.
According to this server device, the genre score can be calculated using the genre score of the query node connected by a node higher than the first location node.

In another preferable aspect, the plurality of nodes include a first location node indicating first location information indicating a location of a document browsed by the user as a result of the search using a hierarchical structure, and more than the first location information. A second location node indicating higher-order second location information, wherein the relay node is the first location node, and the calculation means includes the second location information, the genre, and the second location information for each second location information. A parameter indicating a correlation between the base node and the process using the genre score of the query indicated by another query node related to the second location node related to the base node and the parameter, and the query of the base node The genre score may be calculated.
According to this server device, the genre score can be calculated using the parameter indicating the correlation between the second location information and the genre.

In yet another preferred aspect, the server device is further provided with third acquisition means for acquiring a search log in which the query used for the search and the search count that is the number of times the search has been performed, and the query are input. Then, to the search server that outputs the search result, a query included in the search log is input, a fourth acquisition unit that acquires the search result for the query, the search log acquired by the third acquisition unit, and the search log Estimation means for estimating the weight of the edge using the search result acquired by the fourth acquisition means, and the processing is indicated by another query node related to the second location node related to the base node Processing for calculating the genre score of the query of the base node may be performed using the genre score of the query and the estimated weight.
According to this server device, the genre score can be calculated from the search log in which the query used for the search and the number of searches that is the number of times the search has been performed are accumulated.

Further, the present invention stores a search log including a plurality of items including a query used for search and first location information indicating a location of the document browsed by the search in a hierarchical structure in the first storage unit. A step of extracting, as a node, items other than the query and the query among items included in the search log stored in the first storage unit; and a query node indicating the extracted query; Generating a graph connected to the relay node indicating the extracted other item by an edge, a query, a genre identifier indicating the genre of the query, and a genre indicating the degree of relevance between the query and the genre The genre score and the query corresponding to the designated genre identifier are acquired from the second storage means storing the genre dictionary including the score. A node identified from among the target nodes satisfying a predetermined condition among the query nodes included in the generated graph as a base node, and the other connected by an edge via the base node and the relay node The process of calculating the genre score of the query of the base node by using the genre score of the query indicated by the query node from the acquired genre scores while updating the base node until the end condition is satisfied A genre score calculation method is provided that includes repeatedly performing genre score calculations for a plurality of queries.
According to this method, a genre score can be calculated using a genre score of another query.

In a preferred aspect, the plurality of nodes are a first location node indicating first location information indicating a location of a document browsed by the user as a result of the search using a hierarchical structure, and a higher rank than the first location information. 2 including a second location node indicating location information and a user node indicating the name of the searched user, wherein the relay node is the first location node, and the processing is connected to the relay node at an edge The genre score of the base query may be calculated using the genre score of a query indicated by the second location node or another query node connected to the user node by an edge.
According to this method, the genre score can be calculated using the genre score of the query node connected by a node higher than the first location node.

In another preferable aspect, the plurality of nodes include a first location node indicating first location information indicating a location of a document browsed by the user as a result of the search using a hierarchical structure, and more than the first location information. Including a second location node indicating higher-order second location information, wherein the relay node is the first location node, and for each second location information, a correlation between the second location information and the genre is calculated. And calculating the parameter to indicate, wherein the processing uses the genre score of the query indicated by the other query node related to the second location node related to the base node and the parameter to use the query of the base node The genre score may be calculated.
According to this method, the genre score can be calculated using the parameter indicating the correlation between the second location information and the genre.

Furthermore, the present invention includes a step of storing a search log including a query used for a search and the number of times the search has been performed in a first storage unit, and a search server that outputs a search result when the query is input On the other hand, a step of inputting a query included in the search log stored in the first storage unit, obtaining a search result for the query, and an item included in the search log stored in the first storage unit A step of extracting the query and other items other than the query as nodes, a query node indicating the extracted query, and a graph connecting the relay node indicating the extracted other item with an edge Generating, using the search log stored in the first storage means and the acquired search result to estimate the weight of the edge, A genre score corresponding to the specified genre identifier from the second storage means storing a genre identifier including a genre identifier indicating the genre of the query and a genre score indicating the degree of relevance between the query and the genre And a step of obtaining a query, and an edge through the base node and the relay node with a node identified from target nodes satisfying a predetermined condition among query nodes included in the generated graph as a base node A process of calculating the genre score of the query of the base node using the genre score of the query indicated by the other query node connected in the above from the acquired genre score and using the estimated weight Repeatedly while updating the base point node until the end condition is satisfied, Providing genre score calculation method and a step of calculating a genre score of the query.
According to this method, the genre score can be calculated from the search log in which the query used for the search and the search count that is the number of times the search is performed are accumulated.

Further, according to the present invention, a search log including a plurality of items including a query used for a search and first location information indicating a location of a document browsed by the search in a hierarchical structure is stored in a computer. And a step of extracting, as nodes, items other than the query and the query among items included in the search log stored in the first storage unit, and the extracted query. A query node, a step of generating a graph connected to the relay node indicating the extracted other item by an edge, a query, a genre identifier indicating the genre of the query, and a high degree of relevance between the query and the genre The genre score corresponding to the specified genre identifier is stored in the second storage means storing the genre dictionary including the genre score indicating And obtaining a query, and using the node specified from among the target nodes satisfying a predetermined condition among the query nodes included in the generated graph as a base node, an edge through the base node and the relay node The process of calculating the genre score of the query of the base node using the genre score of the query indicated by the other query node connected in the above-described genre score until the end condition is satisfied. A program for executing a step of calculating genre scores of a plurality of queries is repeatedly performed while updating nodes.
According to this program, a genre score can be calculated using a genre score of another query.

Furthermore, the present invention stores a search log including a query used for a search and the number of times the search is performed in a first storage means, and outputs a search result when the query is input. A query included in the search log stored in the first storage means and obtaining a search result for the query; and a search log stored in the first storage means. Among the included items, the step of extracting the query and other items other than the query as nodes, the query node indicating the extracted query, and the relay node indicating the extracted other item are connected at the edge A weight of the edge is estimated using a step of generating the graph, a search log stored in the first storage means, and the acquired search result Corresponding to the specified genre identifier from the second storage means storing the step, the query, the genre identifier indicating the genre of the query, and the genre score indicating the degree of relevance between the query and the genre A genre score to be obtained and a query, and a node identified from target nodes satisfying a predetermined condition among query nodes included in the generated graph as a base node, and the base node and the relay node A genre score of a query indicated by another query node connected by an edge is used from among the acquired genre scores, and the genre score of the query of the base node is determined using the estimated weight. The calculation process is repeated while updating the base node until the end condition is satisfied. Returns conducted to provide a program for executing the steps of calculating a genre score multiple queries.
According to this program, a genre score can be calculated using a genre score of another query.

It is a figure showing composition of search system 1 concerning one embodiment. It is a figure which illustrates a search log. It is a figure which illustrates a graph. It is a figure explaining the production | generation process of a graph. It is a figure which illustrates a genre dictionary. 2 is a diagram illustrating a hardware configuration of a genre server 10. FIG. It is a flowchart which shows a genre score calculation process. It is a figure which shows the graph acquired in this example. It is a figure which shows the accompanying data of the graph of FIG. It is a flowchart which shows an example of a genre score calculation process. It is a figure which illustrates the output data from genre server. It is a figure which shows the graph as a proportionality. It is a figure which illustrates the graph concerning the modification 1. It is a figure which shows the data accompanying the graph of FIG. 10 is a flowchart showing an operation according to Modification 2. It is a figure which illustrates the graph concerning the modification 2. It is a figure which shows the data accompanying the graph of FIG. It is a figure which illustrates the graph concerning the modification 3. It is a figure which shows the structure of the search system 1 which concerns on the modification 4. It is a figure which illustrates the search log concerning the modification 4. It is a figure which illustrates a search result. It is a figure which illustrates the table containing the calculated estimated value.

1. Configuration FIG. 1 is a diagram illustrating a configuration of a search system 1 according to an embodiment. The search system 1 includes a genre server 10, a search server 20, and a client 30. In the search system 1, when a genre identifier is input from the client 30, the genre server 10 outputs a genre score of a query corresponding to the genre identifier.

  The genre server 10 includes, as functional elements, a storage unit 11, a storage unit 13, a generation unit 12 (an example of a third acquisition unit, a fourth acquisition unit, and an estimation unit), a calling unit 14, and a calculation unit 15 (first acquisition unit). , A second acquisition unit and an example of a calculation unit) and an output unit 16. The storage unit 11 stores a search log. The search log is a log in which data of a plurality of items including a search query and an identifier of a document browsed as a result of the search are accumulated.

  FIG. 2 is a diagram illustrating a search log. In this example, the search log includes a plurality of data sets (records). Each data set includes data divided into a plurality of items. As items, a user name, a query, a URL (Uniform Resource Locator), a time stamp, an order, a first place URL, a second place URL, and a third place URL are used. The searched document is a document existing on the Internet. Here, the “document” refers to an HTML (HyperText Markup Language) file, an audio file, an image file, and other types of files. The location of the document is indicated by a URL (an example of first location information) that is location information using a hierarchical structure. The user name indicates the user who performed the search. The query indicates a word used for searching the document. The URL indicates the location of a document viewed (selected) by the user from a plurality of documents extracted by the search. The time stamp indicates the time when the search is performed. The rank indicates the rank in the search result of the document viewed by the user. The first URL indicates the location of the first document in the search result. The same applies to the second URL and the third URL. In the example of FIG. 2, the data set in the top row indicates that the user with the user name “u1” performed a search using the query “q1” on January 25, 2010, 0: 0: 38. Yes. Further, in this search, the documents with URLs “s1”, “s2”, and “s4” are presented as the first to third positions in the search result, and the user selects the document (URL “s1”) from among them. It is shown that the search result is the first document).

  Refer to FIG. 1 again. The generation unit 12 generates a graph from the search log stored in the storage unit 11. Here, the “graph” refers to information representing a connection relationship between a plurality of nodes by an edge (link) or a visual expression thereof.

  FIG. 3 is a diagram illustrating a graph. Circles indicate nodes, and solid lines indicate edges. In this example, a query and a domain (an example of second location information) are used as nodes. Hereinafter, a query node is referred to as a “query node”, and a domain node is referred to as a “domain node” (an example of a second location node). The same applies to other nodes. Looking at the domain as a reference, for example, the query nodes q1 and q2 are connected to the domain node d1, and the query node q3 is connected to the domain node d2. Alternatively, when viewed on the basis of a query, domain nodes d2, d3, and d5 are connected to the query node q3. From this graph, it can be read that a query q1 is input and that a user has viewed a document belonging to the domain d1 is recorded in the search log.

  FIG. 4 is a diagram for explaining a graph generation process. Here, an example of generating a graph from the search log shown in FIG. 2 will be described. In FIG. 4, the upper table corresponds to the search log of FIG. The graph is generated as follows: The generation unit 12 reads the search log from the storage unit 11. The generation unit 12 extracts a node from the read search log. Here, the user name, query, URL, and domain are extracted as nodes. These items are arranged in a hierarchy. The hierarchy has a structure of a user name, a query, a URL, and a domain in order from the bottom. Which items are extracted as nodes and the hierarchical structure of the nodes are determined in advance. The domain is extracted from the URL. The URL has a hierarchical structure described in a predetermined format such as “aaa: //bbb.ccc.ddd/eee/fff”. The domain is a character string from the top of the hierarchical structure, specifically, from the top of the URL to before the first “/” next to “: //”. In this example, “aaa: //bbb.ccc.ddd” is the domain. In this example, the domain is a concept higher in hierarchy than the URL, and a plurality of domains are not extracted from a single URL (that is, a plurality of domains do not correspond to a single URL). The generation unit 12 generates an edge for each data set. In the example of FIG. 4, the first data set includes a user name “u1”, a query “q1”, and a URL “s1”. The domain “d1” is extracted from the URL “s1”. The generation unit 12 connects the node u1, the node q1, the node s1, and the node d1 to the first data set by edges. An edge is provided between nodes in adjacent hierarchies. For example, the user node and the query node are connected at the edge because the hierarchy is adjacent. Since the user node and the URL node (an example of the first location node) are not adjacent in the hierarchy (there is a query node between them), they are not directly connected at the edge. The second data set in FIG. 4 includes a user name “u1”, a query “q1”, and a URL “s2”. The domain “d2” is extracted from the URL “s2”. The generation unit 12 connects the node u1, the node q1, the node s2, and the node d2 to the second data set by edges. Similarly, for all data sets, nodes are connected by edges.

  The generation unit 12 further generates accompanying data accompanying the graph from the search log. The accompanying data includes data used for calculating the genre score. Details of the accompanying data will be described later.

  Refer to FIG. 1 again. The storage unit 13 stores a genre dictionary. The genre dictionary is a database indicating the relationship between queries and genres.

  FIG. 5 is a diagram illustrating a genre dictionary. The genre dictionary includes a plurality of data sets. Each data set includes data divided into a plurality of items. As items, a query, a genre identifier, a genre, and a genre score are used. The genre indicates the classification or type of the document. The genre identifier is a genre identifier. The genre score indicates the degree of relevance between the query and the genre. In this example, the highest score of the genre score is 1.00 and the lowest score is 0.00. For example, when the genre score is 1.00, the query and the genre are highly relevant. The first data set in FIG. 5 indicates that the query “q1” is related to the genre “Comedy” (= genre identifier “1567”), and the genre score is 1.00. The genre score is calculated by PageRank, HITS algorithm, SALSA, TrustRank or other known algorithms. The calculation of the genre score may be performed by the genre server 10 (calculation unit 15), or may be performed by a device other than the genre server 10, such as the search server 20.

  Refer to FIG. 1 again. The calling unit 14 receives a genre identifier from the client 30 as an argument. The calling unit 14 extracts a query corresponding to the received genre identifier from queries included in the genre dictionary stored in the storage unit 13. The calling unit 14 outputs the extracted query to the calculation unit 15.

  The calculation unit 15 calculates a genre score for the query input from the calling unit 14. The output unit 16 outputs the genre score calculated by the calculation unit 15 to the client 30.

  FIG. 6 is a diagram illustrating a hardware configuration of the genre server 10. The genre server 10 includes a control unit 110, a storage unit 120, an input unit 130, a display unit 140, and a communication unit 150. The control unit 110 controls other components. The control unit 110 includes a CPU (Central Processing Unit) 111, a RAM (Random Access Memory) 112, and a ROM (Read Only Memory) 113. The CPU 111 is a device that performs various calculations. The RAM 112 is a storage device that functions as a work area when the CPU 111 executes a program. The ROM 113 is a storage device that stores programs and data. The storage unit 120 includes a nonvolatile storage device such as a built-in flash memory, an HDD (Hard Disk Drive), and a removable memory card. The storage unit 120 stores various programs and data. The control unit 110 executes the genre score calculation program stored in the storage unit 120, thereby realizing the function of FIG. The functions for storing information, such as the storage unit 11 and the storage unit 13, secure a storage area in the RAM 112 and the storage unit 120 and store the information in this storage area.

  The input unit 130 inputs information to the control unit 110. The input unit 130 includes input devices such as a keyboard and a mouse. The display unit 140 displays information. The display unit 140 includes a display device such as an LCD (Liquid Crystal Display) or an EL display. The communication unit 150 performs communication via a network.

  The control unit 110 executing the genre score calculation program is an example of the generation unit 12, the calling unit 14, and the calculation unit 15. The communication unit 150 is an example of the output unit 16. The storage unit 120 cooperating with the control unit 110 is an example of the storage unit 11 and the storage unit 13.

  The search server 20 outputs a search result in response to a search request from another device such as the client 30. The search request includes a query. The search result includes the URL of the document searched using the query and the rank of the document. The search server 20 includes hardware elements (CPU, ROM, RAM, storage unit, input / output unit, etc.) and software for realizing this function (not shown).

  The client 30 is a device operated by a user, for example, a personal computer or a mobile phone. The client 30 transmits a search request including a query input by the user to the search server 20. The client 30 displays the search result transmitted from the search server 20. In addition, the client 30 transmits a genre score calculation request including the genre identifier to the genre server 10. The client 30 proposes a query using the genre score transmitted from the genre server 10. The client 30 includes hardware elements (CPU, ROM, RAM, storage unit, input / output unit, display unit, etc.) and software for realizing this function (not shown).

2. Operation FIG. 7 is a flowchart showing a genre score calculation process in the search system 1. In this example, the storage unit 13 stores the genre dictionary shown in FIG. 5 before the flow of FIG. 7 is started.

  In step S100, the genre server 10 accumulates search logs. That is, the storage unit 11 receives a search log from the search server, and accumulates the received search log in addition to the stored search log. The search log is accumulated at a predetermined timing each time a search is performed in the search server 20, or after a certain period of time has elapsed since the previous accumulation.

  In step S <b> 110, the calling unit 14 acquires a genre identifier from the client 30. The genre identifier is included in the genre score calculation request transmitted from the client 30. Here, a case where the genre identifier acquired by the calling unit 14 is “1567” will be described as an example. In this case, the calling unit 14 extracts a query corresponding to the acquired genre identifier from the data set included in the genre dictionary (FIG. 5) (step S120). In this example, queries q1, q3 and q4 are extracted. The calling unit 14 outputs a data set including the extracted query and its genre score to the calculating unit 15.

In step S <b> 130, the generation unit 12 generates a graph and accompanying data using the search log stored in the storage unit 11.
In step S140, the calculation unit 15 calculates the genre score of the target query by propagating the genre score from other queries using the graph generated by the generation unit 12. Details are as follows. First, the calculation unit 15 acquires a graph from the generation unit 12.

  FIG. 8 is a diagram showing a graph acquired in this example. In this example, the graph includes a query node and a domain node. The query node includes nodes for queries q1, q2, q3, and q4. The domain node includes nodes of domains d1, d2, and d3. The query node q1 is connected to the domain nodes d1 and d2. Query node q2 is connected to domain nodes d1 and d2. The query node q3 is connected to the domain node d1. The query node q4 is connected to the domain nodes d2 and d3.

  FIG. 9 is a diagram showing accompanying data of the graph of FIG. The accompanying data includes a plurality of data sets. Each data set includes data of a query, a domain, the number of unique users (NUU), a unique user rate (RUU), and the number of searches (Search). The number of unique users indicates the number of unique users who have browsed documents belonging to the corresponding domain among the documents searched using the corresponding query. A unique user is a unique number of people who visited a document during a specific period. Even when the same person browses the document five times within the same period, the number of unique users is one. The first data set in FIG. 9 indicates that the number of unique users who browsed documents belonging to the domain d1 among the documents searched using the query q1 is 90. The unique user rate indicates a ratio of unique users who have browsed a document belonging to a corresponding domain among documents searched using a corresponding query. The first data set in FIG. 9 indicates that the ratio of unique users who viewed documents belonging to the domain d1 among documents retrieved using the query q1 is 0.9 (= 90%). Yes. This is the ratio of the unique users (90 people) who have viewed documents belonging to the domain d1 out of all unique users (90 + 10 = 100 people) who have browsed the documents searched using the query q1. The number of searches indicates the number of searches performed using the corresponding query. The first data set in FIG. 9 indicates that 200 searches have been performed using the query q1.

  With reference to FIG. 7 again, the process of step S140 will be described. The calculation unit 15 acquires a data set including a query and a genre score from the calling unit 14. In step S130, the calculation unit 15 performs a process of calculating a genre score for the target query among the queries included in the acquired data set using the accompanying data of the graph. In this example, the calculation unit 15 calculates a genre score for each of the queries q1, q2, and q3.

FIG. 10 is a flowchart illustrating an example of the genre score calculation process in step S140. In step S200, the calculation unit 15 determines initial values of genre scores for all queries included in the graph. The initial value is determined as follows. First, for queries (queries q1, q2, q4) included in the data set output in step S110, the genre score included in this data set is used as an initial value. Next, 0.00 is used as an initial value for a query not included in this data set (query q3), that is, a query whose genre score is not described in the genre dictionary. In this example, the initial value JS _ini of the genre score of the queries q1-q4 is determined as in the following equation (1) based on the genre dictionary of FIG.

  In step S <b> 210, the calculation unit 15 specifies a query (hereinafter referred to as “target query”) that is a target of genre score propagation processing. The calculation unit 15 identifies a query that satisfies a predetermined condition as a target query. In this example, the condition that “the initial value of the genre score is zero” (= the genre score is not recorded in the genre dictionary) is used as a condition for specifying the target query. In the example of Expression (1), only the query q3 is specified as the target of the propagation process, and the queries q1, q2, and q4 are not the targets of the propagation process. Here, “propagation of genre score” means that the genre score of a certain query is calculated using the genre score of another query. For example, calculating the genre score of the query q2 using the genre scores of the queries q1 and q3 is referred to as “propagating the genre scores of the queries q1 and q3 to the query q2.”

  In step S220, the calculation unit 15 identifies a query serving as a base point from the target queries. Hereinafter, a query serving as a base point is referred to as a “base point query”, and a node of the base point query is referred to as a “base point node”. The base point query is determined by a predetermined algorithm. This algorithm is, for example, an algorithm that specifies a query as a base point query one by one from the target query. In this case, the target queries are sorted according to a rule such as an Iueo order, an alphabetical order, a character code order, or an ID number order. In this example, since the target query is only the query q2, the query q2 is specified as the base point query.

In step S230, the calculation unit 15 calculates the edge weight for the route including the base point query. Here, the “path” refers to a connection relationship between two nodes on the graph via edges and nodes. In the example of FIG. 8, the query node q2 is connected to the query nodes q1 and q3 via the domain node. This connection includes the following three paths (A) to (C).
(A) A path A connected to the query node q1 using the domain node d1 as a relay node.
(B) A path B connected to the query node q3 using the domain node d1 as a relay node.
(C) A path C connected to the query node q1 using the domain node d2 as a relay node.
These routes are classified into the following two types depending on the node that is a relay node.
(1) Route group 1 (routes A and B) having the domain node d1 as a relay node.
(2) Route group 2 (route C) using the domain node d2 as a relay node.

ここで、ＲＵＵ_qi,djは、クエリｑｉを用いて検索された文書の中からドメインｄｊに属する文書を閲覧したユニークユーザの割合を示す。 The calculation unit 15 calculates the edge weights P _{qi, dj, qr} of the path connecting the base node qr, the domain node dj as a relay node, and another query node qi by the following equation (2).

Here, RUU _{qi, dj} indicates the percentage of unique users who have browsed documents belonging to the domain dj from among the documents searched using the query qi.

ここで、Ｓｅａｒｃｈ_qiは、クエリｑｉを用いて検索が行われた回数を示す。なお、式（３）においてｑｉはｑｉ＝ｑｒとなる場合を含まない。 In step S240, the calculation unit 15 calculates the genre score JS (qr) of the base point query qr by the following equation (3). That is, the calculation unit 15 propagates the genre scores of other queries to the starting point query.

Here, Search _qi indicates the number of times a search is performed using the query qi. In Expression (3), qi does not include the case where qi = qr.

次に、算出部１５は、式（３）を用いてジャンルスコアＪＳを算出する。ここでは紙面の都合により、経路群１と経路群２を分けて計算する。まず、経路群１については、

である。経路群２については、

である。式（５）および式（６）から、

が得られる。 A calculation example using Expression (2) and Expression (3) will be specifically described with an example in which the base query is the query q2. The calculation unit 15 calculates the edge weight P for each route using Equation (2).

Next, the calculation unit 15 calculates the genre score JS using Expression (3). Here, the route group 1 and the route group 2 are calculated separately due to space limitations. First, for route group 1,

It is. For route group 2,

It is. From Equation (5) and Equation (6),

Is obtained.

  In step S250, the calculation unit 15 determines whether genre score propagation processing has been performed for all target queries using the queries as base queries. If there is a target query that has not yet become the base query (S250: NO), the calculation unit 15 moves the process to step S220. When the propagation process is performed using all target queries as the base query (S250: YES), the calculation unit 15 proceeds to step S260.

  In step S260, the calculation unit 15 determines whether the end condition is satisfied. When it is determined that the end condition is satisfied (S260: YES), the calculation unit 15 ends the process of FIG. When it is determined that the termination condition is not satisfied (S260: NO), the calculation unit 15 shifts the process to step S220 again. In this example, the end condition satisfies at least one of the following conditions: (1) the genre score has converged, and (2) the process of steps S220-250 is repeated a certain number of times (for example, 10 laps). This is the condition.

  For example, when there are ten target queries q1, q2,..., Q10, first, a genre score propagation process is performed using the query q1 as the target query. Next, genre score propagation processing is performed using the query q2 as a target query. Similarly, up to query q10, genre score propagation processing is performed as the target query one by one in order. When the propagation process is completed once for each of the queries q1, q2,..., Q10, it is said that “the propagation process is rotated once”. When the propagation process is performed once, it is determined whether the end condition is satisfied. If the end condition is not satisfied, the propagation process for the second round is performed. Hereinafter, the propagation process is repeated for the third, fourth,... Until the end condition is satisfied.

  Again referring to FIG. In step S <b> 150, the output unit 16 outputs data including the genre score to the client 30. Further, the output unit 16 writes the updated genre score in the genre dictionary.

  FIG. 11 is a diagram illustrating output data from the genre server 10. The output data includes a plurality of data sets. Each data set includes a query and a genre score for the query. The client 30 receives output data from the genre server 10. Thus, the client 30 can obtain a genre score in response to a genre score calculation request. The client 30 displays the queries included in the received output data that satisfy a predetermined condition (for example, the top five genre scores) as recommended queries.

  FIG. 12 is a diagram showing a graph as a comparative example. In this contrast, the graph includes a query node and a URL node. Based on this graph, it is also possible to calculate the genre score using the equations (2) and (3) (in this case, the term relating to the domain node in the equation is replaced with the URL node). However, when the URL node is used, the edge density is coarser than when the domain node is used. This is often due to the fact that documents that can be searched through a search service are sparse. When trying to propagate a genre score from one query to the other for two queries, the genre score cannot be propagated unless the query nodes of these queries are connected to a common URL node. . That is, when the graph of FIG. 12 is used, there is a problem that a genre score cannot be propagated to a query node that is not connected to a common URL node. On the other hand, according to the genre server 10, even if the query node is not connected to the common URL node, the genre score can be propagated to the query node connected to the common domain node. it can. With such a configuration, a graph can be created more flexibly. Further, when the genre score is propagated from a plurality of queries when calculating the genre score, an index indicating the relevance to the plurality of queries can be calculated at a time.

  As another contrast, the contents of the search log may be limited. For example, the number of searches for each query is recorded as a log, but the search behavior history for each user or the browsing rate for each document may not be recorded as a log. Such a limitation is caused by, for example, a lack of hardware or software resources or a search server specification. For example, it is assumed that the browsing rate for each document is recorded as a log, and the technique for calculating the importance for each document using this and further calculating the genre score using the importance is a search log. There is a problem that it cannot be applied to an environment in which the content of this is restricted. On the other hand, according to the genre server 10, the genre score can be propagated even when the search behavior history for each user or the browsing rate for each document is not recorded as a log.

  As described above, according to the genre server 10, a genre score can be calculated even for a query for which a conventional genre score could not be calculated. When a graph including a domain node is used, a genre score can be propagated in a wider range than when a URL node is used.

3. Other Embodiments The present invention is not limited to the above-described embodiments, and various modifications can be made. Hereinafter, some modifications will be described. Two or more of the following modifications may be used in combination.

3-1. Modification 1
FIG. 13 is a diagram illustrating a graph according to the first modification. The structure of the graph generated by the generation unit 12 is not limited to that illustrated in FIG. In the example of FIG. 13, the graph includes a query node, a URL node, and a domain node. These nodes are arranged in a hierarchy. In this example, the query node, URL node, and domain node are arranged in this order from the bottom. In this example, the calculation unit 15 propagates the genre score in consideration of the edges of the query node and the URL node.

  FIG. 14 is a diagram showing data associated with the graph of FIG. This data includes a plurality of data sets. Each data set includes query, URL, domain, NUU, RUU, search count, and page view (PV) data. In this example, the number of unique users indicates the number of unique users who have browsed a document at a corresponding URL among documents retrieved using a corresponding query. The first data set in FIG. 14 indicates that there are 90 unique users who have browsed the document at URLs1 among the documents searched using the query q1. The page view indicates the number of times the document at the corresponding URL has been browsed. The first data set in FIG. 14 indicates that the document at URLs1 has been viewed 80 times.

次に、算出部１５は、式（３）に従ってジャンルスコアＪＳを算出する。 The calculation unit 15 calculates the edge weight P according to the following equation (8) instead of the equation (2).

Next, the calculation unit 15 calculates the genre score JS according to Equation (3).

さらに、式（３）から、ジャンルスコアＪＳは以下のとおり算出される。

Here, a specific example of calculation will be shown by taking the case where the query q2 is a base point query as an example. From the data in FIG. 14 and the equation (2), the edge weights P _{q1, s1, d1, s2, q2} are calculated as follows.

Furthermore, the genre score JS is calculated from the equation (3) as follows.

ここで、係数Ｗｅｉｇｈｔ_djは、ドメインｄｊと対象となるジャンルとの間の相関を示すパラメータ（以下「伝播調整パラメータ」という）である。この例で、係数Ｗｅｉｇｈｔ_djは、相関係数βであり、ジャンルスコアＪＳ、検索回数Ｓｅａｒｃｈ、ユニークユーザ率ＲＵＵの関数である。

以下、式（１１）によるジャンルスコアの算出を、具体例を用いて説明する。 3-2. Modification 2
The formula for calculating the genre score is not limited to the formula (3). Equation (3) may be modified using a weighting factor. In Modification 2, the genre score is calculated by the following equation (11).

Here, the coefficient Weight _dj is a parameter indicating the correlation between the domain dj and the target genre (hereinafter referred to as “propagation adjustment parameter”). In this example, the coefficient Weight _dj is a correlation coefficient β, and is a function of the genre score JS, the search count Search, and the unique user rate RUU.

Hereinafter, the calculation of the genre score by Expression (11) will be described using a specific example.

  FIG. 15 is a flowchart illustrating an operation according to the second modification of the search system 1. The flow of FIG. 15 is different from the flowchart of FIG. 10 in that step S300 is inserted between steps S220 and S230. In step S300, the calculation unit 15 calculates a propagation adjustment parameter for each domain according to Equation (12). In step S240, the calculation unit 15 calculates a genre score according to equation (11). Hereinafter, a specific example of calculation will be described.

  FIG. 16 is a diagram illustrating a graph according to the second modification. The graph of this example has the same hierarchical structure as the graph of FIG.

  FIG. 17 is a diagram showing data associated with the graph of FIG. This data includes a plurality of data sets. Each data set includes query, URL, domain, NUU, RUU, and search count data.

ここでは、Ｗｅｉｇｈｔ_d2＝０．００１であった場合を例に説明する。 In step S300, the calculation unit 15 calculates the propagation adjustment parameter Weight. From the data in FIG. 17 and the equation (12), the propagation adjustment parameter Weight _d2 is calculated as follows.

Here, a case where Weight _d2 = 0.001 will be described as an example.

In step S230, the calculation unit 15 calculates the edge weight P according to the equation (2). Here, a specific example of calculation will be shown by taking the case where the query q2 is a base point query as an example. From the data of FIG. 17 and the equation (2), the edge weights P _{q1 and q2} are calculated as follows.

The calculation unit 15 calculates the genre score JS in step S240. The genre score JS (q2) is calculated from the data of FIG. 17 and the equation (11) as follows.

In another example, when the query node q2 and the query node q4 are connected via the URL node belonging to the domain node d4 in the graph of FIG. 16, the genre score JS (q2) is calculated as follows.

  For example, when there is a document with a low relationship with a specific genre such as Wikipedia (registered trademark), if a graph including only a query node and a URL node is used as in the past, the relationship with the genre is Could not evaluate accurately. For example, when only the graph of FIG. 12 is used and the genre dictionary is not used, it is not easy to determine whether a certain URL (that is, the document itself) belongs to a specific genre. An example is a dictionary site such as Wikipedia. A dictionary site is thought to be tied to queries of various genres, so it may not have a specific single genre. When only the connection between the URL node and the query node is considered, there is a problem that the genre score is propagated through the URL node that does not have a specific genre. According to the genre server 10, the genre score is propagated from the query node associated with the common genre. According to the configuration of this modification, the genre score that is propagated by using the propagation adjustment parameters aggregated for each domain can be suppressed on a domain basis, and a genre score with higher accuracy can be calculated.

3-3. Modification 3
FIG. 18 is a diagram illustrating a graph according to the third modification. In the example of FIG. 18, the graph includes a query node, a URL node, a domain node, and a user node. These nodes are arranged in a hierarchy. In this example, the query node, URL node, domain node, and user node are arranged in this order from the bottom. Domain nodes and user nodes are arranged in the same hierarchy.

  In this example, the genre score is propagated through the URL node and the user node in addition to the route through the domain node. For example, the genre score of the query q1 is propagated to the query q2 via the URL node s2, the user node u2, and the URL node s3. As described in the second modification, the genre score of the query q2 is calculated using a weighting coefficient based on the domain node.

  According to this configuration, since the query node is propagated via the URL node or the user node, a more comprehensive graph is generated as compared with the case where a graph including only the query node and the URL node is used. Is done. That is, it is possible to give a genre score to more queries.

3-4. Modification 4
FIG. 19 is a diagram illustrating a configuration of the search system 1 according to the fourth modification. In this example, the genre server 10 further includes a storage unit 17 in addition to the configuration of FIG. The storage unit 17 stores the graph data generated by the generation unit 12.

  FIG. 20 is a diagram illustrating a search log according to the fourth modification. In the search log of this example, each data set includes data of a query and the number of searches, but does not include data of a user and a viewed document (URL). In the above-described embodiment, the edge weight P is calculated using the number of query searches and the number of unique users who have viewed the document. However, as in this example, when the search log does not include data of the user and the viewed document (URL), the edge weight P cannot be calculated by the same method as in the above-described embodiment. Therefore, in this example, the generation unit 12 generates a graph using data acquired from the search server 20 in addition to the search log stored in the storage unit 11. Details are as follows.

  The generation unit 12 acquires a search log from the storage unit 11. The generation unit 12 extracts a query from the acquired search log. In the example of FIG. 20, queries q1 and q2 are extracted. The generation unit 12 transmits a search request including the extracted query to the search server 20. Upon receiving a search request from the generation unit 12 (genre server 10), the search server 20 performs a search and transmits the result to the genre server 10.

  FIG. 21 is a diagram illustrating search results. In this example, the search result includes a plurality of data sets. Each data set includes query, URL, and ranking data. In the example of FIG. 21, the first to second data sets indicate that the URLs1 document is first and the URLs2 document is second when a search is performed using the query q1.

  The generation unit 12 calculates an estimated value of missing data from the search result. Missing data refers to data that is not included in the search log among the data used to calculate the genre score. For example, when the genre score is calculated according to equations (2) and (3), the unique user rate RUU is missing data. The estimated value of missing data is calculated as follows, for example. Each URL is assigned the number of searches according to the order. For example, the number of searches is assigned at a ratio of first place: second place = 2: 1. In the example of FIG. 20, 158 times, the number of unique users is assigned to URLs1 (first document) and URLs2 (second document) at a ratio of 2: 1. That is, the generation unit 12 assigns 105.33 unique users to URLs1 and 52.67 unique users to URLs2 out of a total of 158 searches performed using the query q1. The generation unit 12 calculates an estimated value of the unique user rate using the number of unique users assigned in this way.

  FIG. 22 is a diagram illustrating a table including the calculated estimated values. The generation unit 12 generates the accompanying data of the graph using the estimated value of the unique user rate calculated as described above.

  According to this configuration, even if the search log does not completely include the data illustrated in FIG. 2, the genre score can be calculated by estimating the missing data.

3-5. Other Modifications The search log may be generated by the genre server 10 itself or by a device other than the genre server 10 (for example, the search server 20). When the genre server 10 generates a search log, the search server 20 transmits a search result to the genre server 10 each time a search is performed. The calculation unit 15 of the genre server 10 adds the received search result to the search log stored in the storage unit 11. When the search server 20 generates a search log, the search server 20 has a storage unit that stores the search log. The search server 20 adds a search result to the search log each time a search is performed. The search server 20 transmits the search log to the genre server 10 at a predetermined timing (after a certain period of time has elapsed since the previous search log was transmitted, or when there is a request from the genre server 10). The storage unit 11 of the genre server 10 stores the received search log.

  The format of the search log is not limited to that illustrated in FIG. For example, in the example of FIG. 2, the URL of the document obtained as a result of the search is recorded from the first place to the third place, but the number of URLs to be recorded is not limited to this. The number of URLs recorded in the search log is limited by, for example, the system configuration (for example, the amount of hardware resources).

  Further, the genre server 10 may not include the storage unit 11. In this case, another device different from the genre server 10 has the storage unit 11. When accessing the search log, the genre server 10 accesses the search log via the other device.

  The method for specifying a genre identifier is not limited to that described in the embodiment. That is, the input to the calling unit 14 is not limited to the genre identifier. A query may be input to the calling unit 14 instead of the genre identifier. In this case, the calling unit 14 refers to the genre dictionary and extracts a query corresponding to the same genre as the genre corresponding to the input query. In the example of FIG. 5, when a query q1 is input, the calling unit 14 extracts queries q3 and q4 corresponding to the same genre as the genre corresponding to the query q1 (“comed”).

  A document to be searched is not limited to a file on the Internet. The genre score calculation technique described in the embodiment may be applied to a file on a local network or a stand-alone computer. For example, when this genre score calculation technique is applied to a file on a stand-alone computer, all of the functions as the genre server 10, the search server 20, and the client 30 have a single computer. The URL is read as a file name including a drive name and a folder (directory) name. The domain is replaced with a folder name at a higher level.

  The method for extracting the domain from the URL is not limited to that described in the embodiment. For example, a character string from the beginning of the URL to the second “/” next to “: //” may be extracted as a domain. In this case, when the URL is “aaa: //bbb.ccc.ddd/eee/fff”, the character string “aaa: //bbb.ccc.ddd/eee” is extracted as a domain.

  The data structure of the genre dictionary is not limited to that shown in FIG. For example, either one of the genre identifier and the genre may be omitted. Alternatively, a plurality of genres may be associated with one query. In this case, in response to a query associated with a plurality of genres, the calling unit 14 extracts only a data set including the genre that is the target in step S120.

  The input source of the genre identifier (genre score calculation request) and the output device of the genre score may not be the same. For example, the genre server 10 may output the genre score to a device different from the client 30 that is the input source of the genre identifier. Alternatively, the genre server 10 may output a genre score to the client 30 for a genre identifier input from a device different from the client 30.

  The data structure of the graph is not limited to that described in the embodiment. Only the table described as the accompanying data in the embodiment may be processed as a graph.

  The conditions for specifying the target query are not limited to those described in the embodiment. In the embodiment, the calculation unit 15 uses a query whose genre score initial value is zero as a target query, but may use a query whose genre score initial value is a predetermined threshold value or less as a target query. Alternatively, the calculation unit 15 may not calculate the initial value of the genre score, and may simply use a query whose genre score is not described in the genre dictionary as a target query. In yet another example, the calculation unit 15 may specify a query specified by the user as a target query.

  A node serving as a relay node is not limited to a domain node. Nodes other than domain nodes, such as user nodes and URL nodes, may be relay nodes.

The algorithm used for calculating the edge weight (that is, the strength of connection between nodes) is not limited to that described in the embodiment. For example, a well-known algorithm such as PageRank, HITS algorithm, SALSA, or TrustRank used for the structure analysis of the hyperlink may be used.
Further, the edge weight may be calculated together with the generation of the graph. That is, in the flow of FIG. 10, the process of step S230 may be outside the process loop of steps S220-S250.

  The flow of the genre score calculation process is not limited to that shown in FIG. For example, the writing of the genre dictionary in step S150, that is, the updating of the genre dictionary may be omitted. In another example, the order of the process of step S120 and the process of step S130 may be switched.

  A device that transmits a genre score calculation request (genre identifier) may be different from a device that receives a response to the request.

  The method of using the genre score in the client 30 is not limited to query recommendation. Processing other than query recommendation, such as processing for displaying a genre score or processing for recommending a genre, may be performed.

  The hardware configuration for realizing the functions shown in FIG. 1 is not limited to that described in FIG. For example, instead of a general-purpose CPU, a processor that performs specific processing may be used.

  In the above-described embodiment, the program executed by the CPU 111 includes a magnetic recording medium (magnetic tape, magnetic disk (HDD, FD (Flexible Disk)), etc.), an optical recording medium (optical disk (CD (Compact Disk)), DVD (Digital Versatile). Disk)), etc.), a magneto-optical recording medium, and a computer-readable recording medium such as a semiconductor memory (flash ROM or the like). The program may be downloaded via a network such as the Internet.

DESCRIPTION OF SYMBOLS 1 ... Search system, 10 ... Genre server, 11 ... Storage part, 12 ... Generation part, 13 ... Storage part, 14 ... Calling part, 15 ... Calculation part, 16 ... Output part, 17 ... Storage part, 20 ... Search server, DESCRIPTION OF SYMBOLS 30 ... Client, 110 ... Control part, 111 ... CPU, 112 ... RAM, 113 ... ROM, 120 ... Storage part, 130 ... Input part, 140 ... Display part, 150 ... Communication part

Claims (10)

First acquisition means for acquiring a graph in which a plurality of nodes including a query node indicating a query used for search and a relay node indicating other items related to the search other than the query are connected by an edge;
A genre score and a query corresponding to a specified genre identifier are acquired from a genre dictionary including a query, a genre identifier indicating the genre of the query, and a genre score indicating the degree of relevance between the query and the genre. A second acquisition means;
Among the query nodes included in the graph acquired by the first acquisition means, a node identified from target nodes indicating queries for which a genre score is not described in the genre dictionary is used as a base node, and the base node and the The genre score of the query of the base node is calculated using the genre score of the query indicated by the other query node connected at the edge via the relay node from the genre score acquired by the second acquisition unit. A server apparatus comprising: calculation means for repeatedly performing the process while updating the base node until an end condition is satisfied, and calculating genre scores of a plurality of queries. The plurality of nodes include a first location node indicating first location information indicating a location of a document browsed by the user as a result of the search using a hierarchical structure, and second location information higher than the first location information. A second location node to indicate, and a user node indicating the name of the user who performed the search,
The relay node is the first location node;
The process calculates a genre score of the base query using a genre score of a query indicated by the second location node connected to the relay node at the edge or another query node connected to the user node at the edge. The server device according to claim 1, wherein the server device performs processing. A first location node indicating first location information indicating a location of a document browsed by the user as a result of the search using a hierarchical structure, and second location information higher than the first location information A second location node indicating
The relay node is the first location node;
The calculation means calculates, for each second location information, a parameter indicating a correlation between the second location information and the genre,
The process is a process of calculating a genre score of the query of the base node using the genre score of the query indicated by another query node related to the second location node related to the base node and the parameter. The server device according to claim 1, characterized in that: Third acquisition means for acquiring a search log in which the query used for the search and the number of searches that are the number of times the search has been performed are accumulated;
A fourth acquisition unit that inputs a query included in the search log to a search server that outputs a search result when a query is input, and acquires the search result for the query;
An estimation unit that estimates the weight of the edge using the search log acquired by the third acquisition unit and the search result acquired by the fourth acquisition unit;
The process is a process of calculating a genre score of a query of the base node using a genre score of a query indicated by another query node related to the second location node related to the base node and the estimated weight. The server device according to claim 1, wherein: The control means of the computer stores a search log including a plurality of items including the query used for the search and the first location information indicating the location of the document browsed by the search in a hierarchical structure in the first storage means. Steps,
The step of the control means extracting the query and other items other than the query among the items included in the search log stored in the first storage means as nodes;
The control means generating a graph connected to the query node indicating the extracted query and the relay node indicating the other extracted items by an edge;
The genre score corresponding to the specified genre identifier from the second storage means storing the query, the genre identifier indicating the genre of the query, and the genre dictionary including the genre score indicating the relevance between the query and the genre And the control means obtains the query;
Other query nodes connected to the base node by an edge via the relay node, with a node specified from target nodes satisfying a predetermined condition among the query nodes included in the generated graph as a base node The process of calculating the genre score of the query of the base node using the genre score of the query indicated by is performed from among the acquired genre scores while updating the base node until an end condition is satisfied, A genre score calculation method comprising: calculating genre scores of a plurality of queries by the control means . The plurality of nodes include a first location node indicating first location information indicating a location of a document browsed by the user as a result of the search using a hierarchical structure, and second location information higher than the first location information. A second location node to indicate, and a user node indicating the name of the user who performed the search,
The relay node is the first location node;
The process uses the genre score of the base query using the genre score of the query indicated by the second location node connected to the relay node at the edge or another query node connected to the user node at the edge, The genre score calculation method according to claim 5, wherein the control means calculates the genre score. A first location node indicating first location information indicating a location of a document browsed by the user as a result of the search using a hierarchical structure, and second location information higher than the first location information A second location node indicating
The relay node is the first location node;
For each of the second location information, the control means further includes a step of calculating a parameter indicating a correlation between the second location information and the genre,
In the processing, the control means calculates the genre score of the query of the base node using the genre score of the query indicated by another query node related to the second location node related to the base node and the parameter. It is processing. The genre score calculation method of Claim 5 characterized by the above-mentioned. Storing the search log including the query used for the search and the number of times the search has been performed, in the computer storage means in the first storage means;
A step of inputting a query included in a search log stored in the first storage unit to a search server that outputs a search result when a query is input, and the control unit acquires a search result for the query When,
The step of the control means extracting the query and other items other than the query among the items included in the search log stored in the first storage means as nodes;
The control means generating a graph connected to the query node indicating the extracted query and the relay node indicating the other extracted items by an edge;
The control means estimating the weight of the edge using the search log stored in the first storage means and the acquired search results;
The genre score corresponding to the specified genre identifier from the second storage means storing the query, the genre identifier indicating the genre of the query, and the genre dictionary including the genre score indicating the degree of relevance between the query and the genre And the control means obtains the query;
Other query nodes connected to the base node by an edge via the relay node, with a node specified from target nodes satisfying a predetermined condition among the query nodes included in the generated graph as a base node The processing of calculating the genre score of the query of the base node using the genre score of the query indicated by is used from the acquired genre scores and using the estimated weights until the end condition is satisfied A genre score calculation method comprising: repeatedly performing genre scores of a plurality of queries by the control means while updating the base node. On the computer,
Storing a search log including a plurality of items including a query used for the search and first location information indicating a location of the document browsed by the search in a hierarchical structure in the first storage unit;
Of the items included in the search log stored in the first storage unit, extracting the query and other items other than the query as nodes;
Generating a graph connected to the query node indicating the extracted query and the relay node indicating the other extracted items by an edge;
The genre score corresponding to the specified genre identifier from the second storage means storing the query, the genre identifier indicating the genre of the query, and the genre dictionary including the genre score indicating the relevance between the query and the genre And retrieving a query,
Other query nodes connected to the base node by an edge via the relay node, with a node specified from target nodes satisfying a predetermined condition among the query nodes included in the generated graph as a base node The process of calculating the genre score of the query of the base node using the genre score of the query indicated by is performed from among the acquired genre scores while updating the base node until an end condition is satisfied, A program for executing a step of calculating a genre score of a plurality of queries. On the computer,
Storing a search log including the query used for the search and the number of times the search has been performed in the first storage unit;
Inputting a query contained in a search log stored in the first storage means to a search server that outputs a search result when a query is input, and obtaining a search result for the query;
Of the items included in the search log stored in the first storage unit, extracting the query and other items other than the query as nodes;
Generating a graph connected to the query node indicating the extracted query and the relay node indicating the other extracted items by an edge;
Estimating the weight of the edge using the search log stored in the first storage means and the acquired search result;
The genre score corresponding to the specified genre identifier from the second storage means storing the query, the genre identifier indicating the genre of the query, and the genre dictionary including the genre score indicating the degree of relevance between the query and the genre And retrieving a query,
Other query nodes connected to the base node by an edge via the relay node, with a node specified from target nodes satisfying a predetermined condition among the query nodes included in the generated graph as a base node The processing of calculating the genre score of the query of the base node using the genre score of the query indicated by is used from the acquired genre scores and using the estimated weights until the end condition is satisfied A program for repeatedly executing the genre score of a plurality of queries by updating the base node.

Images