JP4848388B2 - How to calculate a score for a search query - Google Patents

Description

  The present invention relates to a method and apparatus for calculating a score related to a search query.

  Conventionally, in a search service on the Web, a keyword input by a user is used as a query, and a Web page including this keyword is searched. Then, a method of displaying the URL of the Web page extracted as a search result on a screen as a list is used. At this time, in order to lead to a desired Web page more efficiently, the search results are often ranked based on a predetermined index indicating the attention level and search frequency of the Web page.

Further, in order to lead to a Web page desired by the user or highly necessary for the user, a method using a past browsing history including other users has also been proposed. For example, in Patent Document 1, the operation history for the Web page of the user group and the purchase history of the product of EC (Electronic Commerce) site are stored in the server, and the product is purchased when there is a request specifying the name of the product from the user. It is shown that a Web page browsed by the users of the selected user group is extracted.
JP 2004-326537 A

  However, the method disclosed in Patent Document 1 is still the same as searching for a Web page that matches a keyword input by the user. Therefore, unless an appropriate keyword is input, it is not possible to easily reach the Web page desired by the user. That is, there is a problem that the accuracy of the query is required because the reliability of the search result largely depends on the query input by the user.

  Therefore, in order to arrive at the Web page desired by the user, it is necessary to input a new effective query, which is preferably provided as recommendation information. Accordingly, an object of the present invention is to provide a method of automatically calculating a score for evaluating a new query or URL that is a candidate for recommendation information according to a user's search intention for an input search query. To do.

  The present invention provides the following solutions.

(1) A method for calculating a score related to a query that is an input to a search engine,
A storage step of storing history data including a query log including a keyword as the query, a URL as a search result by the search engine, a rank of the URL, and click-through data regarding the URL;
In order to generate a graph structure of a query definition in which the nodes are connected by a URL common to a plurality of nodes among URLs browsed based on a search result corresponding to the query of the node, using the query as a node. And an analysis step for analyzing the history data;
An extraction step of extracting a combination of a recommendation source query and a recommendation destination query connected by a URL from the graph structure;
A calculation step for calculating a score for the combination extracted by the extraction step based on the click-through data and the rank data;
Associating at least one combination extracted by the extracting step with one recommendation source query.

  According to such a configuration, the server executing the method includes a query log including a keyword as a query, a URL as a search result by a search engine, and a rank of the URL, and click-through data regarding the URL ( History data including the number of clicks of the URL and the ratio of the number of clicks with respect to the number of times the URL is displayed, the query as a node, and the URL browsed based on the search result according to the query of the node, History data is analyzed in order to generate a graph structure of a query definition in which the nodes are connected by a URL common to a plurality of nodes. Then, a combination of the recommendation source query and the recommendation destination query connected by the URL is extracted from the graph structure, a score for the extracted combination is calculated based on the click-through data and the rank data, and at least one extracted Are associated with one recommendation source query.

  Thus, the server can generate a graph structure in which a query is a node and nodes are connected by a common URL based on historical data including a query log and click-through data. The score based on the click-through data and the rank data is calculated for the combination of the recommendation source query and the recommendation destination query extracted from the graph structure, and is associated with one recommendation source query. The score of each recommendation destination query can be obtained and evaluated.

(2) accepting an input query and identifying the recommendation source query;
The method according to (1), further comprising an output step of outputting at least one recommended query from a combination extracted in association with the input query in response to receiving the input query.

  According to such a configuration, the server that executes the method accepts the input query, specifies the recommendation source query, and in response to receiving the input query, from the combination extracted in association with the input query. , Output at least one recommendation destination query.

  Accordingly, the server can output at least one recommendation destination query when the input query is a recommendation source query in response to receiving the input query. Therefore, a recommendation destination query different from the input query can be presented according to the score calculated based on the history data.

(3) A method of calculating a score for a query that is an input from a user to a search engine in a terminal device, a search server including a search engine, and a server connected via a network. ,
A query input to the search engine in the terminal device, a URL browsed based on a search result of the search engine according to the input of the query, and a rank of the browsed URL in the search result An accumulation step for accumulating as history data in association with each other;
Based on the accumulated history data, out of a plurality of queries associated with the same URL, a query having a higher evaluation value of each query based on the rank is set as a recommendation destination query, and a query other than the recommendation destination query is set as a recommendation source. An extraction step for extracting a combination of the recommendation source query, the URL, and the recommendation destination query as a query;
By analyzing the input / output relationship of the edge when each query is a node of the graph structure in the graph structure constituted by the set of the extracted combinations and connecting a plurality of queries via the URL, the user A calculation step of calculating a score of each of the queries input from.

  According to such a configuration, the server executing the method includes the query input to the search engine in the terminal device, the URL browsed based on the search result of the search engine according to the input of the query, and the The ranks of the browsed URLs in the search results are accumulated as history data in association with each other, and based on the accumulated history data, among the multiple queries associated with the same URL, A query having a higher evaluation value is set as a recommendation destination query, a query other than the recommendation destination query is set as a recommendation source query, and a combination of the recommendation source query, the URL, and the recommendation destination query is extracted. Then, in a graph structure that is composed of a set of combinations and a plurality of queries are connected via a URL, the input / output relationship of the edge when analyzing each query as a node of the graph structure is input from the user. Calculate the score for each query that is executed.

  Accordingly, the server extracts a recommendation destination query and a URL for a certain query based on history data (click log) related to a URL searched and browsed by the user in the past, and a plurality of queries are connected through the URL. A graph structure can be constructed. Then, by analyzing the input / output relationship of the edge that nodes each query in this graph structure, that is, the link relationship of the URL, a score indicating the degree of attention of each query is calculated. A score corresponding to the user's search intention can be calculated.

  That is, the server applies the analysis method used for the static hyperlink structure on the Internet based on the input / output relationship of the edge to each node in the graph structure to the dynamic data called click log. Can do.

  The above score can be calculated by applying existing analysis methods such as PageRank (registered trademark), HITS, SALSA, etc. (for example, “mining the web-Discovering Knowledge from Hypertext Data” (Soumen Chakrabarti, Morgan Kaufmann Publishers, 2003) ”). These methods are analysis methods of hyperlink structures on the Internet, but the above graph structure can also be applied because it is a structure in which queries are linked by URLs.

  Further, a plurality of types of scores may be calculated by adopting these plurality of analysis methods. Furthermore, these multiple types of scores may be integrated by a method such as obtaining a weighted average, for example, and another evaluation value may be obtained.

(4) In a terminal device, a search server provided with a search engine, and a server connected via a network, a score for a URL related to a query that is an input from a user to the search engine is calculated. A method,
A query input to the search engine in the terminal device, a URL browsed based on a search result of the search engine according to the input of the query, and a rank of the browsed URL in the search result An accumulation step for accumulating as history data in association with each other;
Based on the accumulated history data, out of a plurality of queries associated with the same URL, a query having a higher evaluation value of each query based on the rank is set as a recommendation destination query, and a query other than the recommendation destination query is set as a recommendation source. An extraction step for extracting a combination of the recommendation source query, the URL, and the recommendation destination query as a query;
By analyzing the input / output relationship of edges when each URL is a node of the graph structure in a graph structure composed of a set of the extracted combinations and connecting a plurality of URLs via a query, the URL A calculation step of calculating each score.

  According to such a configuration, the server that executes the method extracts a combination of a recommendation source query, a URL, and a recommendation destination query in the same manner as in (3). Then, in the graph structure composed of this set of combinations and connecting a plurality of URLs via a query, by analyzing the input / output relationship of the edge when each URL is a node of the graph structure, each URL is analyzed. Calculate the score.

  As a result, the server extracts a recommendation destination query and a URL for a query based on history data (click log) related to a URL searched and browsed by the user in the past, and a plurality of URLs are connected through the query. A graph structure can be constructed. Then, by analyzing the input / output relationship of the edge that noded the URL in this graph structure, that is, the search relationship with the query, a score indicating the degree of attention of each URL is calculated. A score corresponding to the user's search intention can be calculated.

  That is, the server applies the analysis method used for the static hyperlink structure on the Internet based on the input / output relationship of the edge to each node in the graph structure to the dynamic data called click log. Can do.

  In addition, said score can be calculated by applying the existing analysis method similarly to (3). A plurality of analysis methods may be employed to calculate a plurality of types of scores. Furthermore, these multiple types of scores may be integrated by a method such as obtaining a weighted average, for example, and another evaluation value may be obtained.

  (5) In response to a new query input from the terminal device, a query related to the newly input query is extracted as recommendation information based on the graph structure and the score, and the terminal device The method according to (3), further including a first transmission step of transmitting to.

  According to such a configuration, the server that executes the method can extract a query as recommendation information based on the graph structure and the calculated score and present it to the user.

  In this way, by recommending a new query based on the calculated score, it is possible to effectively recommend a query with a high degree of attention in a search executed in the past, so that the user can easily follow a desired Web page. You can expect to arrive.

  Here, as the query output as recommendation information, a query located at a lower level in the graph structure, that is, a recommendation destination query for the recommendation source query may be extracted as a candidate, but is not limited thereto. For example, a query located at the top, that is, a recommendation source query for a recommendation destination query may be extracted as a candidate. Furthermore, it is good also as tracing a graph structure regardless of a high-order or a low-order, and giving priority to the near query by the evaluation value according to the distance. Further, a query in the vicinity of a query located at a lower level may be output as a lower concept, and a query near a query located at a higher level may be output as a higher concept.

  Thus, there is a possibility that the server can provide effective recommendation information by setting an evaluation method according to the situation.

  (6) In response to a new query input from the terminal device, a URL related to the newly input query is extracted as recommendation information based on the graph structure and the score, and the terminal device The method according to (4), further including a first transmission step of transmitting to

  According to such a configuration, the server that executes the method can extract the URL as the recommendation information based on the graph structure and the calculated score and present it to the user.

  In this way, by recommending a new URL based on the calculated score, it is possible to effectively recommend a URL with a high degree of attention in a search executed in the past, so that the user can easily follow a desired Web page. You can expect to arrive.

  Here, similar to (5), the URL to be output as the recommendation information is a URL positioned at a lower level in the graph structure, that is, a URL positioned between the recommendation source query and the recommendation destination query as candidates. It may be extracted, but is not limited to this. For example, a URL that is positioned higher, that is, a URL that is positioned between the recommendation destination query and the recommendation source query may be extracted as a candidate. Furthermore, the graph structure may be traced regardless of whether it is higher or lower, and a nearby URL may be prioritized by an evaluation value corresponding to the distance. Alternatively, a URL in the vicinity of a lower URL may be output as a lower concept, and a URL in the vicinity of a higher URL may be output as a higher concept.

  Thus, there is a possibility that the server can provide effective recommendation information by setting an evaluation method according to the situation.

  (7) The method according to (5), wherein the first transmission step ranks and extracts a query whose score is within a predetermined value with respect to the newly input query. .

  According to such a configuration, the server executing the method can preferentially recommend a query whose score matches or approximates the input query. Here, since the queries whose scores match or approximate are often synonyms and synonyms, there is a possibility that a query that can acquire similar information can be effectively presented.

  (8) The first transmission step is characterized in that, among the recommendation information, those whose scores are within a predetermined value are grouped and extracted, according to any one of (5) to (7) the method of.

  According to such a configuration, a server that executes the method can group queries whose scores match or approximate to each other. Thus, for example, regarding synonyms and synonyms, it is expected that the user can easily recognize the recommended information by displaying only the representative or distinguishing it from another group.

  Further, the server can group URLs whose scores match or approximate each other. Thus, for example, only representatives of URLs for different notations for the same Web page, URLs for Web pages with similar contents, and the like can be displayed separately from different groups.

  (9) The first transmission step calculates an evaluation value of each of the recommended information for the newly input query based on the score, and excludes evaluation values that are less than a predetermined value. The method according to any one of (5) to (8), wherein the extraction is performed.

  According to such a configuration, a server that executes the method can evaluate a query or URL that is a candidate for recommendation information with respect to a newly input query. If the evaluation is low, it is possible that effective recommendation information can be presented to the user by excluding it from the recommendation information.

  (10) The method further includes a second transmission step of transmitting a search result by the search engine based on the newly input query when the recommendation information is not extracted in the first transmission step. To (9).

  According to such a configuration, the server executing the method can present a URL related to the input query by a conventional search method when there is no information to be recommended.

  (11) In the first transmission step, a query having a predetermined similarity or higher with the newly input query is selected from the graph structure, and the recommendation information is selected based on the selected query. The method according to any one of (5) to (10), wherein the extraction is performed.

  According to such a configuration, when extracting recommendation information, the server that executes the method selects data that matches or is similar to the input query from the prepared graph structure.

  As a result, the server selects a query that is assumed to be synonymous from the graph structure based on the similarity measured by the partial matching or the edit distance of the character string, as well as the completely matching query. Can do. Therefore, there is a possibility that the fluctuation of the query input by the user can be absorbed and processed effectively as the same query.

(12) A score calculation device that is connected to a terminal device and a search server including a search engine via a network and calculates a score for a query that is an input from a user to the search engine,
A query input to the search engine in the terminal device, a URL browsed based on a search result of the search engine according to the input of the query, and a rank of the browsed URL in the search result Storage means for storing the history data in association with each other;
Based on the accumulated history data, out of a plurality of queries associated with the same URL, a query having a higher evaluation value of each query based on the rank is set as a recommendation destination query, and a query other than the recommendation destination query is set as a recommendation source. An extraction unit that extracts a combination of the recommendation source query, the URL, and the recommendation destination query as a query;
By analyzing the input / output relationship of the edge when each query is a node of the graph structure in the graph structure constituted by the set of the extracted combinations and connecting a plurality of queries via the URL, the user Calculating means for calculating the score of each query input from,
A score calculation device comprising:

  According to such a configuration, the same effect as in (3) can be expected by operating the score calculation apparatus.

(13) A score calculation device that is connected to a terminal device and a search server including a search engine via a network and calculates a score for a URL related to a query that is an input from a user to the search engine. Because
A query input to the search engine in the terminal device, a URL browsed based on a search result of the search engine according to the input of the query, and a rank of the browsed URL in the search result Storage means for storing the history data in association with each other;
Based on the accumulated history data, out of a plurality of queries associated with the same URL, a query having a higher evaluation value of each query based on the rank is set as a recommendation destination query, and a query other than the recommendation destination query is set as a recommendation source. An extraction unit that extracts a combination of the recommendation source query, the URL, and the recommendation destination query as a query;
By analyzing the input / output relationship of edges when each URL is a node of the graph structure in a graph structure composed of a set of the extracted combinations and connecting a plurality of URLs via a query, the URL A score calculation device comprising: calculation means for calculating each score.

  According to such a configuration, the same effect as in (4) can be expected by operating the score calculation apparatus.

  According to the present invention, it is possible to automatically calculate a score that evaluates a new query or URL that is a candidate for recommendation information according to a user's search intention for an input search query.

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

[System configuration]
FIG. 1 is a block diagram showing a search system according to an example of a preferred embodiment of the present invention.

  The recommendation server 10, the search server 20, the content server 30, and the terminal device 40 are connected via a network. The user of the terminal device 40 accesses the search server 20, and acquires a search result by inputting a query (search keyword) for reaching a desired Web page to a predetermined search engine. The user selects a URL listed in the search result and browses a Web page managed by the content server 30.

  The recommendation server 10 accumulates history data (click data) such as a URL viewed by a user based on a search result in response to a query input to a search engine of the search server 20. The recommendation server 10 determines a recommendation query or recommendation URL as recommendation information to be newly recommended in association with the input query, based on the score indicating the degree of attention based on the history, and transmits the recommendation query or the recommendation URL to the terminal device 40. The user of the terminal device 40 arrives at a desired Web page by executing a new search using a recommended query that is different from the query input by the user or by following a recommended URL.

  Here, in order for the recommendation server 10 to determine the recommendation information, first, based on the click data, graph data in which the query and the URL are continuously connected is generated and accumulated, and the query included in the graph data and A score for evaluating the attention degree of the URL is calculated. Thereafter, when a new query is received from the terminal device 40, a query or URL to be recommended is determined based on the calculated score. Details of these processes will be described below.

  In addition, although the recommendation server 10 demonstrates as a single server, it is not restricted to this, You may distribute to a some server suitably for every below-mentioned various functions.

[Function block]
FIG. 2 is a block diagram showing a functional configuration of the recommendation server 10 according to an example of the preferred embodiment of the present invention.

  The recommendation server 10 includes a click data acquisition unit 110, a click data storage unit 115, a graph creation unit 120, a graph data storage unit 125, a query score calculation unit 130, a synonymous query extraction unit 140, a URL score calculation unit 150, and a score storage unit 155. A query receiving unit 160, a search processing unit 170, a recommendation data display unit 180, and a search engine calling unit 190.

  The click data acquisition unit 110 acquires click data from the search server 20 and stores it as history data in the click data storage unit. The click data is stored accumulatively, for example, as a click data table shown in FIG. 3, and includes at least a query, a rank, and a URL.

  The query included in the click data is a character string input by the user of the terminal device 40 to the search engine included in the search server 20. The URL of the click data indicates the URL clicked by the user in the list of URLs obtained as a search result by the search engine. The rank indicates the rank of the clicked URL in the search result list. For example, the rank corresponds to the rank from the top of the search result list.

  The graph creation unit 120 reads the click data stored in the click data storage unit 115, creates the graph data, and stores it in the graph data storage unit 125. Details of the process for creating the graph data will be described later (FIG. 10).

  The created graph data is stored, for example, as a query definition graph data table shown in FIG. 4 or a URL definition graph data table shown in FIG. In FIG. 4, the graph data includes a second query as a recommendation destination query for the first query (recommendation source query), and is stored together with a URL that can be searched from both the first query and the second query. The recommendation destination query is a query that ranks higher than the search performed by the recommendation source query in order to obtain the URL associated with the recommendation source query and the recommendation destination query as a search result.

  Also, in FIG. 4, the first query and the second query that are the same in the query definition graph data table so that the second query (□□) in the first row is the same as the first query in the second row. Are associated with each other to create a URL definition graph data table storing graph data in which the first URL (recommendation source URL), the second URL (recommendation destination URL) and the query in FIG. 5 are associated with each other.

  Here, the set of query definition graph data shown in FIG. 4 can search a recommended query (q2) recommended for a recommendation source query (for example, q1) from both queries as shown in FIG. The graph structure of the query definition linked through a simple URL (u1) is constructed.

  Further, as shown in FIG. 7, the set of URL definition graph data shown in FIG. 5 is searched by the URL (u1) searched by the recommendation destination query (for example, q4) and further by the recommendation destination query ( The URL-defined graph structure is configured by linking u2) via the recommendation destination query (q2).

  As shown in FIG. 6, a graph in which a query is noded is called a “query definition” graph, and a graph in which a URL is a node as shown in FIG. 7 is called a “URL definition” graph. These two types of graph structures are not limited to the tree type as shown in FIG. 6, and often have a structure in which links are looped, similar to a hyperlink structure on the Internet.

  The query score calculation unit 130 acquires query definition graph data (FIG. 4) from the graph data storage unit 125, and calculates a score (query score) indicating the attention level of each query. As a score to be calculated, a page rank (registered trademark) score, an authority score by a HITS (Hypertext Induced Topic Selection) algorithm, and a hub score are adopted by an existing method for analyzing a hyperlink structure on the Internet. These are techniques for calculating the score of a node based on the input / output relationship of edges in the graph structure. According to the graph definition of the query definition, the score of the query as the node is calculated based on the input / output relationship of the URL as the edge. Note that any other score may be used as long as it is a method for calculating the score of a node in the graph structure.

  The URL score calculation unit 150 obtains URL-defined graph data (FIG. 5) from the graph data storage unit 125, and calculates a score (URL score) indicating the attention level of each URL. The calculation method is similar to the query score. According to the URL definition graph definition, the URL score as a node is calculated based on the input / output relationship of the query as an edge. At this time, the score calculated by the query score calculation unit 130 is input as an initial parameter for analysis.

  The URL score calculated in this way is further input as an initial parameter of the query score calculation unit 130, and the recommendation server 10 repeats the calculation of the query score and the URL score. Thus, the query and URL scores converge, and the converged values are stored in the score storage unit 155 as, for example, a query definition score table shown in FIG. 8 and a URL definition score table shown in FIG. In FIG. 8, the score for the second query (recommendation destination query) is stored, and in FIG. 9, the score for the second URL (recommendation destination URL) is stored.

  The query receiving unit 160 newly receives a search query from the terminal device 40 and passes it to the search processing unit 170.

  The search processing unit 170 refers to the score storage unit 155 for the query received from the query reception unit 160, and searches for a query or URL as recommendation information based on the stored score.

  The recommendation data display unit 180 transmits a query or URL as recommendation information searched by the search processing unit 170 to the terminal device 40 and displays it as a search result. The user of the terminal device 40 can further perform a search or browse the recommended URL based on the displayed recommendation information.

  The search engine calling unit 190 calls the search engine of the search server 20 or another search engine when the recommended information is not searched in the search processing unit 170, and based on the query received by the query receiving unit 160, The URL search is executed by the method described above. Thereby, the recommendation server 10 can provide a search result to a user even when recommendation information cannot be output.

[Graph data creation processing]
FIG. 10 is a flowchart of processing for creating graph data according to an example of the preferred embodiment of the present invention.

  In step S <b> 11, the graph creation unit 120 reads click data from the click data storage unit 115.

  In step S12, the graph creation unit 120 extracts a recommendation query, that is, a query that can easily reach each URL, based on the read click data.

  Specifically, records holding the same URL are collected from the click data and sorted in descending order of evaluation values based on rank or click frequency. This evaluation value is calculated as, for example, “log (click frequency) / rank”. According to this calculation formula, a large evaluation value is obtained when the click frequency is high and the rank is high (the numerical value is small). As a result, it is possible to determine that a query located at the top of the sorted record is easier to reach a predetermined URL than a query located at the bottom. To do.

  The click frequency may be the number of times the user clicked on the URL of the search result, but is not limited to this. For example, the click frequency may be the ratio of the number of clicks to the number of searches (number of sessions) by the same query.

  FIG. 11 is a table showing the relationship between the extracted recommendation destination query and a query (recommendation source query) that serves as a trigger for recommending the recommendation destination query. According to FIG. 11, the same URL is searched for both by the recommendation destination query and the recommendation source query, and among them, the query having the highest evaluation value based on the click frequency (number of clicks / number of sessions) and rank. Is the recommended query. In this case, if any of the first to third recommendation source queries is input as a query, the corresponding recommendation destination query is set as a recommendation information candidate.

  In step S13, the graph creation unit 120 extracts graph data based on the extracted recommendation destination query.

  Specifically, a combination of a recommendation source query, a URL, and a recommendation destination query is extracted as graph data based on the correspondence relationship of the data table shown in FIG. FIG. 12 is a table showing combinations of extracted graph data. The recommendation source query and the recommendation destination query are stored, and are associated through a URL that is commonly searched from the recommendation source query and the recommendation destination query.

  Note that the combination to be extracted may have an evaluation value based on the above rank or click frequency equal to or higher than a predetermined value, but is not limited to this. For example, this evaluation value is higher than the same recommendation source query. A predetermined number may be extracted.

  In step S <b> 14, the graph creation unit 120 stores the extracted graph data in the graph data storage unit 125. Specifically, as described above, the query definition graph data and the URL definition graph data are stored as the tables shown in FIGS. 4 and 5. Each table may include the rank, the number of sessions, and the number of clicks shown in FIG. 12, and may include the above-described evaluation values based on these.

[Score data creation process]
FIG. 13 is a flowchart showing processing for creating score data according to an example of the preferred embodiment of the present invention.

  In step S <b> 21, the query score calculation unit 130 calculates a score indicating the attention level of each query based on the query definition graph data stored in the graph data storage unit 125. This score can be calculated by using various methods (for example, page rank (registered trademark) or HITS), but for the connection between a query and a URL, attention of the URL in the search by the query Perform weighting that reflects the degree. Further, this weighting may be given a bias regarding the ranking of URLs presented to the user. That is, since it is rare for the user to click on the URL displayed in the lower order of the search result, the lower the position, the higher the weight may be given.

  In step S <b> 22, the URL score calculation unit 150 calculates a score indicating the degree of attention of each URL by using a hyperlink structure analysis method with respect to the graph relation of the URL definition graph data stored in the graph data storage unit 125. . At this time, the query score calculated in step S21 is used as an initial parameter for calculation.

  Subsequently, the URL score calculated in step S22 is input as an initial parameter in step S21, and steps S21 to S22 are repeated. As a result, the query score and the URL score converge to a constant value.

  In step S23, it is determined whether or not the query score calculated in step S21 and step S22 and the URL score have converged to a certain value. If it has been determined that it has converged (determined as YES), the calculation step ends and the process proceeds to step S24. If it has been determined that it has not yet converged (determined as NO), the process returns to step S21. Repeat the score calculation.

  In step S24, the score of the query and the URL score determined to have converged in step S23 are stored in the score storage unit 155.

[Search execution processing]
FIG. 14 is a flowchart showing search execution processing according to an example of the preferred embodiment of the present invention.

  In step S31, the query receiving unit 160 receives a new query for search from the terminal device 40, and the search processing unit 170 executes search processing based on this query.

  In step S32, the search processing unit 170 determines whether the request from the terminal device 40 accompanying the query is a search for a recommendation destination query or recommendation destination URL as recommendation data. If YES is determined, the search for recommendation data is requested, and the process proceeds to step S33. On the other hand, if NO is determined, the search for the related URL with respect to the received query is requested instead of the search for the recommendation data, and the process proceeds to step S37.

  In step S33, the search processing unit 170 extracts a recommendation destination query or a recommendation destination URL as recommendation data based on the received query. At this time, the recommendation data may be evaluated based on the score stored in the score storage unit 155, and the recommendation data having the higher score may be extracted. Examples of recommendation data to be extracted are shown in the following (1) to (7).

  (1) In the query definition graph data (FIG. 4), a recommendation destination query for the same recommendation source query as the received query is extracted as recommendation data. At this time, ranking is performed based on the score of the recommendation destination query. Thereby, the recommendation server 10 can recommend to the user a low-level concept query that can search the URL as a high-rank search result in order to reach the URL desired by the user with respect to the received query.

  (2) In the query definition graph data (FIG. 4), a recommendation source query for the same recommendation destination query as the received query is extracted as recommendation data. At this time, ranking is performed based on the score of the recommendation source query. Thereby, the recommendation server 10 can recommend to the user a high-level concept query that may be able to search the URL as a search result in order to reach the URL desired by the user for the received query.

  (3) A query of a node located in the vicinity of the received query node on the graph structure is extracted as recommendation data. Here, the neighborhood refers to, for example, a query on nodes up to two nodes via a URL that is an edge of the graph structure. Thereby, the recommendation server 10 may be able to recommend a query highly relevant to the received query.

  (4) In the query definition graph data (FIG. 4), a recommendation destination query for the same recommendation source query as the received query is extracted, and a query located in the vicinity of the recommendation destination query on the graph structure is extracted as recommendation data. Thereby, the recommendation server 10 may be able to recommend a query related to the subordinate concept of the received query.

  (5) In the query definition graph data (FIG. 4), a recommendation source query for the same recommendation destination query as the received query is extracted, and a query located in the vicinity of the recommendation source query on the graph structure is extracted as recommendation data. Thereby, the recommendation server 10 may be able to recommend a query related to the superordinate concept of the received query.

  (6) In the query definition graph data or URL definition graph data, a URL located in the vicinity of the received query is extracted as recommendation data. Thereby, the recommendation server 10 may be able to recommend a URL that is related to the received query and has a high degree of attention.

  (7) Corresponding to a query recommended by the methods (1) to (5), a URL associated with the query (a URL serving as a basis for a recommended query) is extracted as recommendation data. Thereby, the recommendation server 10 may be able to present a URL with a high degree of attention together with the recommended query.

  In the recommendation data extraction method as described above, the matching between the received query and the graph data is desirably a perfect match of the character strings, but is not limited thereto. For example, the determination may be made based on matching of partial character strings decomposed by morphological analysis or similarity measured by the edit distance of the character strings.

  The evaluation of the recommendation data based on the score may be the score itself stored in the score storage unit 155, but is not limited thereto. For example, in order to obtain a relative evaluation with respect to the received query (base point), a weighted average obtained by adding a distance from the base point, a URL click frequency, and the like as elements may be calculated as an evaluation value.

  Note that if the evaluation value is lowered as the distance from the base point increases, a query or URL that is close to the received query, that is, has a high relevance, is extracted with priority. Further, by increasing the evaluation value based on the click frequency, there is a possibility that recommendation data with a higher degree of attention is given priority.

  In addition, a query having a score close to that of the received query, that is, a query having a score within a predetermined range may be preferentially extracted. Thereby, recommendation data with similar contents is preferentially presented, and there is a possibility that information related to the received query can be efficiently searched.

  After extracting the recommendation data in step S33 as described above, in step S34, the search processing unit 170 determines whether the recommendation data has been extracted. If YES is determined, the recommendation data has been extracted, and the process proceeds to step S35. If NO is determined, the process proceeds to step S37 because there is no recommendation data.

  In step S35, the recommendation data display unit 180 transmits the recommendation data extracted in step S33 to the terminal device 40 for display. 15 and 16 are diagrams showing examples of screens displaying recommendation data.

  In FIG. 15, as a first example, a recommended query is displayed as a “recommended keyword” for the query “car” input by the user.

  In FIG. 16, as a second example, the recommendation destination query ranked for the query “car” input by the user is displayed as a “recommendation query”. And about each displayed recommendation destination query, URL (URL linked with a recommendation destination query) used as the basis of recommendation is also displayed. FIG. 16 also shows an evaluation value based on the score of the recommendation destination query and the click frequency of the URL.

  Furthermore, the recommendation destination queries are displayed as a group of synonymous queries whose scores are close to each other. Here, for synonymous queries, only those representative queries may be displayed, thereby reducing the number of displayed items and improving the visibility of the user.

  In step S36, the query receiving unit 160 accepts the selection of a query as a request for further search from the displayed recommendation data, and returns to step S32. Specifically, for example, when a “recommended query” in FIG. 16 is clicked, a related URL search request is accepted. In addition, when “more recommended” is clicked on the recommendation query, a recommendation data search request with the corresponding recommendation destination query as the recommendation source query is accepted.

  In step S37, the search engine calling unit 190 calls a conventional search engine such as the search engine of the search server 20, and searches for a URL related to the query. As a result, a URL search based on the recommendation data displayed as the recommendation data by the recommendation server 10 can be executed.

[Hardware configuration of server]
FIG. 17 is a diagram illustrating an example of a hardware configuration of the recommendation server 10 according to an example of the preferred embodiment of the present invention. The recommendation server 10 includes a CPU (Central Processing Unit) 1010 (a plurality of CPUs such as the CPU 1012 may be added in a multiprocessor configuration) constituting the control unit 101 that implements each function of FIG. 2, a bus line 1005, a communication I / F 1040, main memory 1050, BIOS (Basic Input Output System) 1060, USB port 1090, I / O controller 1070, keyboard and mouse 1100, and other input means and display device 1022.

  Storage means such as a tape drive 1072, a hard disk 1074, an optical disk drive 1076, and a semiconductor memory 1078 can be connected to the I / O controller 1070.

  The BIOS 1060 stores a boot program executed by the CPU 1010 when the recommendation server 10 is activated, a program depending on the hardware of the recommendation server 10, and the like.

  The hard disk 1074 constituting the storage unit 107 stores various programs for the recommendation server 10 to function as a server and programs for executing the functions of the present invention, and various databases (for example, click data storage) as necessary. Unit 115, graph data storage unit 125, score storage unit 155, etc.).

  As the optical disc drive 1076, for example, a DVD-ROM drive, a CD-ROM drive, a DVD-RAM drive, or a CD-RAM drive can be used. In this case, the optical disk 1077 corresponding to each drive is used. A program or data can be read from the optical disk 1077 by the optical disk drive 1076 and provided to the main memory 1050 or the hard disk 1074 via the I / O controller 1070.

  The program provided to the recommendation server 10 is provided by being stored in a recording medium such as the hard disk 1074, the optical disk 1077, or a memory card. This program may be installed in the recommendation server 10 and executed by being read from the recording medium via the I / O controller 1070 or downloaded via the communication I / F 1040.

  The aforementioned program may be stored in an internal or external storage medium. Here, as a storage medium constituting the storage unit 107, in addition to the hard disk 1074, the optical disk 1077, or the memory card, a magneto-optical recording medium such as an MD or a tape medium can be used. Further, a storage device such as a hard disk 1074 or an optical disk library provided in a server system connected to a dedicated communication line or the Internet may be used as a recording medium, and the program may be provided to the recommendation server 10 via the communication line.

  Here, the display device 1022 displays a screen for accepting data input to the user or displays a screen of a calculation processing result by the recommendation server 10, and includes a cathode ray tube display device (CRT), a liquid crystal display device ( LCD).

  Here, the input means accepts input by the user, and may be configured by a keyboard, a mouse 1100, and the like.

  The communication I / F 1040 is a network adapter for enabling the recommendation server 10 to be connected to a terminal via a dedicated network or a public network. The communication I / F 1040 may include a modem, a cable modem, and an Ethernet (registered trademark) adapter.

  In the above example, the recommendation server 10 has been mainly described. However, the functions described above can also be realized by installing a program in a computer and operating the computer as a server device. Therefore, the function realized by the recommendation server 10 described as an embodiment in the present invention can be realized by executing the above-described method by the computer, or by introducing the above-described program into the computer and executing it. It is feasible.

[Hardware configuration of terminal device]
The terminal device 40 also has a configuration similar to that of the recommendation server 10 described above. In the above-described example, an example realized by a so-called computer has been described. Furthermore, as long as the principle of the present invention is applicable, it can be realized by various terminals such as a mobile phone, a PDA (Personal Data Assistant), and a game machine. It's okay.

  As mentioned above, although embodiment of this invention was described, this invention is not restricted to embodiment mentioned above. The effects described in the embodiments of the present invention are only the most preferable effects resulting from the present invention, and the effects of the present invention are limited to those described in the embodiments of the present invention. is not.

It is a block diagram which shows the search system which concerns on an example of suitable embodiment of this invention. It is a block diagram which shows the function structure of the recommendation server 10 which concerns on an example of suitable embodiment of this invention. It is a figure which shows the click data table which concerns on an example of suitable embodiment of this invention. It is a figure which shows the query definition graph data table which concerns on an example of suitable embodiment of this invention. It is a figure which shows the URL definition graph data table which concerns on an example of suitable embodiment of this invention. It is a figure which shows the graph structure of the query definition which concerns on an example of suitable embodiment of this invention. It is a figure which shows the graph structure of URL definition which concerns on an example of suitable embodiment of this invention. It is a figure which shows the query definition score table which concerns on an example of suitable embodiment of this invention. It is a figure which shows the URL definition score table which concerns on an example of suitable embodiment of this invention. It is a flowchart of the process which produces the graph data based on an example of suitable embodiment of this invention. It is a figure which shows the relationship between the recommendation destination query which concerns on an example of suitable embodiment of this invention, and a recommendation origin query. It is a figure which shows the combination extracted as graph data based on an example of suitable embodiment of this invention. It is a flowchart which shows the process which produces the score data based on an example of suitable embodiment of this invention. It is a flowchart which shows the process of the search execution which concerns on an example of suitable embodiment of this invention. It is a figure which shows the 1st example of the screen which displayed the recommendation data based on an example of suitable embodiment of this invention. It is a figure which shows the 2nd example of the screen which displayed the recommendation data based on an example of suitable embodiment of this invention. It is a figure which shows an example of the hardware constitutions of the recommendation server 10 which concerns on an example of suitable embodiment of this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 Recommendation server 20 Search server 30 Content server 40 Terminal device 110 Click data acquisition part 115 Click data storage part 120 Graph preparation part 125 Graph data storage part 130 Query score calculation part 150 Score calculation part 155 Score storage part 160 Query reception part 170 Search Processing unit 180 Recommended data display unit 190 Search engine call unit

Claims (13)

A search server with the terminal device, the and search engine, connected via a network server, a method for calculating the score with respect to the input a is a query from a user to the search engine ,
A query input to the search engine in the terminal device, a URL browsed based on a search result of the search engine according to the input of the query, and a rank of the browsed URL in the search result An accumulation step for accumulating as history data in association with each other;
Based on the accumulated history data, out of a plurality of queries associated with the same URL, a query having a higher evaluation value of each query whose value is calculated higher as the rank is higher is set as a recommended query, and the recommendation An extraction step for extracting a combination of the recommendation source query, the URL, and the recommendation destination query using a query other than the destination query as a recommendation source query;
By analyzing the input / output relationship of the edge when each query is a node of the graph structure in the graph structure constituted by the set of the extracted combinations and connecting a plurality of queries via the URL, the user A calculation step of calculating a score of each of the queries input from the server . Terminal, and the search servers with and search engine, connected via a network server calculates a score for the URL associated with the query is input from a user to the search engine A method,
A query input to the search engine in the terminal device, a URL browsed based on a search result of the search engine according to the input of the query, and a rank of the browsed URL in the search result An accumulation step for accumulating as history data in association with each other;
Based on the accumulated history data, out of a plurality of queries associated with the same URL, a query having a higher evaluation value of each query whose value is calculated higher as the rank is higher is set as a recommended query, and the recommendation An extraction step for extracting a combination of the recommendation source query, the URL, and the recommendation destination query using a query other than the destination query as a recommendation source query;
By analyzing the input / output relationship of edges when each URL is a node of the graph structure in a graph structure composed of a set of the extracted combinations and connecting a plurality of URLs via a query, the URL A method in which the server executes a calculation step of calculating each score. In response to a new query input from the terminal device, a query related to the newly input query is extracted as recommendation information based on the graph structure and the score, and transmitted to the terminal device. The method of claim 1 , wherein the server further performs a first sending step. In response to a new query input from the terminal device, a URL related to the newly input query is extracted as recommendation information based on the graph structure and the score and transmitted to the terminal device. The method of claim 2 , wherein the server further performs a first sending step. The first transmission step, among the recommendation information, a method according to any one of the preceding claims 3, wherein the score those within a predetermined value from each other, extracted grouped. The first transmission step calculates an evaluation value of each of the recommended information for the newly input query based on the score, and extracts the evaluation values that do not satisfy a predetermined value. A method according to any one of claims 3 to 5 , characterized in that The server further executes a second transmission step of transmitting a search result by the search engine based on the newly input query when the recommendation information is not extracted in the first transmission step. The method according to any one of claims 3 to 6 . In the first transmission step, a query having a predetermined similarity or higher with the newly input query is selected from the graph structure, and the recommendation information is extracted based on the selected query. A method according to any of claims 3 to 7 , characterized in that The accumulation step further accumulates click-through data including the click frequency of the URL,
The extraction step is characterized in that a query having a higher evaluation value of each query whose value is calculated to be larger as the rank is higher and the click frequency is higher is the recommended query. The method according to claim 8. A method in which a server calculates a score for a query that is input to a search engine,
A storage step of storing history data including a query log including a keyword as the query, a URL as a search result by the search engine, a rank of the URL, and click-through data regarding the URL;
By analyzing the history data , the nodes are connected by a URL common to a plurality of nodes among URLs browsed based on a search result according to the query of the node, with the query as a node. and analysis step that generates the graph structure of the query definition,
An extraction step of extracting a combination of a plurality of queries connected by a URL from the graph structure;
Based on the click-through data and the query log , the score for each query of the combination extracted in the extracting step increases as the click frequency in the click-through data increases and the rank in the query log increases. a calculation step of calculating as,
The server executes an associating step of associating at least the recommendation destination query having the highest score among the queries included in the combination extracted by the extracting step with the recommendation source query having a lower score than the recommendation destination query. how to. And accepting step of accepting an input query,
In response to receiving the input query, the server further executes an output step of outputting at least one recommendation destination query from the recommendation destination query associated with the input query as the recommendation source query. The method of claim 10 . Terminal, and the search servers with and search engines, are connected via a network, a score calculating device for calculating a score against the input query from a user to the search engine,
A query input to the search engine in the terminal device, a URL browsed based on a search result of the search engine according to the input of the query, and a rank of the browsed URL in the search result Storage means for storing the history data in association with each other;
Based on the accumulated history data, out of a plurality of queries associated with the same URL, a query having a higher evaluation value of each query whose value is calculated higher as the rank is higher is set as a recommended query, and the recommendation An extraction means for extracting a combination of the recommendation source query, the URL, and the recommendation destination query, using a query other than the destination query as a recommendation source query;
By analyzing the input / output relationship of the edge when each query is a node of the graph structure in the graph structure constituted by the set of the extracted combinations and connecting a plurality of queries via the URL, the user A score calculation device comprising: a calculation means for calculating a score of each query input from Terminal, and the search engine search servers with, connected via a network, the score calculation unit for calculating a score of relative URL associated with the query is input from a user to the search engine Because
A query input to the search engine in the terminal device, a URL browsed based on a search result of the search engine according to the input of the query, and a rank of the browsed URL in the search result Storage means for storing the history data in association with each other;
Based on the accumulated history data, out of a plurality of queries associated with the same URL, a query having a higher evaluation value of each query whose value is calculated higher as the rank is higher is set as a recommended query, and the recommendation An extraction means for extracting a combination of the recommendation source query, the URL, and the recommendation destination query, using a query other than the destination query as a recommendation source query;
By analyzing the input / output relationship of edges when each URL is a node of the graph structure in a graph structure composed of a set of the extracted combinations and connecting a plurality of URLs via a query, the URL A score calculation device comprising: calculation means for calculating each score.

Images