JP5292250B2 - Document search apparatus, document search method, and document search program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To analyze retrieval intention in detail for each user from a retrieval log to present a highly precise retrieval result. <P>SOLUTION: A query frequency counting part 112 counts a retrieval frequency of each query in a log by referring to a retrieval log DB 110. A retrieval intention analyzing part 116 obtains retrieval intention time series data extracting user retrieval intention expressed in a highly frequently searched query for each optional period. A retrieval intention time series analyzing part 120 determines periodicity of retrieval intention based on data and stores the periodicity in a retrieval intention periodicity DB 122. A query collating part 130 collates an input query of a user terminal 2 with the stored data in the DB 122, and if there is the periodicity, obtains retrieval intention data. A search engine 131 searches an index DB 133 based on the input query. A ranking processing part 135 ranks the retrieval result again by using the retrieval intention data. <P>COPYRIGHT: (C)2011,JPO&amp;INPIT

Description

  The present invention relates to a document search technique for searching an electronic document group based on a search term (query) input by a user, acquiring a search result, and responding to the user.

  2. Description of the Related Art As is well known, a document search system that inputs a search word (query) to a server computer via a client computer connected via the Internet and obtains a search result is currently used by many users. This server computer stores a search log from time to time, and information on the input query and information about a click on the search result remain in the log.

  The document search system of Non-Patent Document 1 uses the log of a document search system typified by a search engine to clarify the search intention of a query, performs a search using the information, and realizes an improvement in search accuracy. ing. This document search system analyzes search engine log information (click log), especially for search result clicks, identifies frequently clicked search results, and uses keywords extracted from their titles and snippets. The query has been expanded.

  This document search technology is based on the user's search act of determining whether or not to click the search result by its title and snippet, and the expanded query complements the search intention of the query with the keyword. It has become. At this time, the click log includes the query and the URL of the search result clicked on it together with the time and user identifier information. However, in the document search system of Non-Patent Document 1, the URL of the search result clicked on the query is clicked. Only the information of is used.

Masaya Murata, Hiroyuki Toda, Yumiko Matsuura, Ryoji Kataoka “Improving Mobile Web-IR Using Access Concentration Sites in Search Research.” of WISE2008 Gui-Rong Xue, Hua-Jun Zeng, Zheng Chen, Yong Yu, Wei-Ying Ma, WenSi Xi, WeiGuo Fan “Optimizing Web Search Using Web Click-Through” Michal Vlachos, Chris Meek, Zogorafula Vagena “Identifying Similiarities, Periodicities and Bursts for Online Search Queries” SIGMOD 2004.

  In Non-Patent Document 1, a method of analyzing a click log, extending a query using a title of a search result that is often clicked and a keyword extracted from a snippet, and complementing the search intention of the input query with the keyword Has proposed.

  However, the search technique of Non-Patent Document 1 simply pays attention to the number of clicks for each page without using information on which user clicked when analyzing the click log. The analysis of each search behavior is not performed.

  As a result, the difference or total number of search intentions of each user expressed by the input query is not known, which may result in insufficient search accuracy. In other words, even if the same query is input, the search intent differs depending on each input user. Therefore, in order to grasp the search intent of the query with higher accuracy, which user clicked the click log It is necessary to analyze using the information of whether or not and analyze each search behavior.

  The present invention has been made to solve the above-mentioned situation, and grasps the difference / total number of search intentions inherent in the query in the search log and their characteristics, and presents a highly accurate search result. Is a solution issue.

  Therefore, the present invention grasps the search intention inherent in the query by clustering the search behavior of the user, that is, the first search behavior and the last search behavior from the search log, and extracting the characteristics of the cluster. At this time, if the search intention expressed by the query changes with time (time), it is preferable to determine the periodicity of the search intention. The search intention grasped here is reflected in the search result.

  One aspect of the document search apparatus of the present invention is a document search apparatus that analyzes a user's search intention expressed by a query in a search log and reflects the result in a search result of an electronic document group, and is included in the search log. Query frequency aggregation means for counting the search frequency of each query for each user based on the number of input queries, and a query set of higher frequencies in the search frequency counted by the query frequency aggregation means as an analysis target, A search intention analysis unit that clusters pairs of the first search action and the last search action and acquires the search intention of the analysis target query represented by each cluster is provided.

  The document search apparatus further includes search intention analysis means for determining the presence or absence of periodicity of the search intention extracted by the search intention analysis means, and if there is periodicity in the search intention, it is reflected in the search result. Also good.

  According to another aspect of the document search apparatus, the periodicity of a user's search intention expressed by a query in a search log is determined, and input to a user terminal using a database storing the search intention having periodicity A document search apparatus that reflects the input time of a received query in a search result, and compares the input query with data stored in the database to check the periodicity of the search intention of the input query, and confirms the periodicity If possible, a query matching unit that acquires stored data corresponding to the input timing of the input query as search intention data, and a search result obtained by searching the electronic document group based on the input query is based on the search intention data acquired by the matching unit. Ranking processing means for performing re-ranking.

  Still another aspect of the document search device is a document search device that searches an electronic document group based on a query input to a user terminal and obtains a search result for responding to the user terminal. A query frequency totaling unit that refers to a search log database that records a search log and counts the search frequency of each query based on the number of times the query is included in the search log, and a higher rank in the search frequency counted by the query frequency totaling unit The query set of frequency is set as the analysis target, and pairs of the first search action and the last search action among the search actions for each user are clustered, and the search intention of the analysis target query represented by each cluster is set for each arbitrary period. Search intention analysis means for obtaining the extracted search intention time series data, and a search result in the search intention time series data extracted by the search intention analysis means. Search intention analysis means for determining the periodicity and storing it in the periodicity database as search intention periodicity data if there is periodicity, and checking the input query by comparing the input query with the data stored in the periodicity database Check the periodicity of the intention, and if the periodicity can be confirmed, a query matching means for acquiring search intention periodicity data corresponding to the input timing of the input query as search intention data, and the search intention data acquired by the query matching means Ranking processing means for reranking the search result based on the search result, and the user click information for the input query and the final result is recorded in the search log database as needed.

  One aspect of the document search method of the present invention is a document search method that analyzes a user's search intention expressed by a query in a search log and reflects the result in a search result of an electronic document group. A query frequency counting step that counts the search frequency of each query for each user based on the number of times the query is included in the search log, and a search intent analysis means that queries the higher frequency in the search frequency counted by the query frequency counting step A search intention analysis step for clustering a set of first search behavior and last search behavior among search behaviors for each user, and obtaining a search intention of an analysis target query represented by each cluster; Have.

  The document search method further includes a search intention analysis step in which search intention analysis means determines whether or not the search intention extracted by the search intention analysis step has periodicity, and if the search intention has periodicity, It may be reflected in the search result.

  According to another aspect of the document search method, the periodicity of a user's search intention expressed by a query in a search log is determined, and input to a user terminal using a database storing the search intention having a periodicity A document search method for reflecting the input time of a received query in a search result, wherein query matching means checks the periodicity of the search intent of the input query by comparing the input query with data stored in the database. If the periodicity can be confirmed, a query collating step for obtaining stored data corresponding to the input time of the input query as search intention data, and a ranking processing means, the search result obtained by searching the electronic document group based on the input query, And a ranking process step for reranking based on the search intention data acquired by the matching means.

  Still another aspect of the document search method is a document search method for searching a group of electronic documents based on a query input to a user terminal and obtaining a search result for responding to the user terminal, the query frequency counting A query frequency counting step in which the means refers to a search log database that records a search log for each user and counts the search frequency of each query based on the number of times the query is included in the search log; A query set having a higher frequency in the search frequency counted in the query frequency counting step is set as an analysis target, and pairs of the first search action and the last search action among the search actions for each user are clustered, and are represented by the respective clusters. A search intention analysis step for acquiring search intention time-series data obtained by extracting the search intention of the analyzed query for each arbitrary period; A search intention analyzing step for determining periodicity in the search intention time-series data extracted in the search intention analysis step, and storing the periodicity in the periodicity database as search intention periodic data if there is periodicity, and query matching The means confirms the periodicity of the search intention of the input query by comparing the input query with the data stored in the periodicity database, and if the periodicity can be confirmed, the search intention periodicity according to the input timing of the input query Query matching step for acquiring data as search intention data, ranking processing means for ranking processing step for reranking the search results based on the search intention data acquired in the query matching step, and for the input query and the final result Record user click information in the search log database as needed ; And a recording step.

  The present invention may be in the form of a program that causes a computer to function as each of the document search devices, and this program may be recorded in a recording medium.

  According to the present invention, the total number of different intentions inherent in the query in the search log and the characteristics thereof can be grasped, and the search accuracy is improved. In particular, by finding a periodic change in search intention, a highly accurate search result is presented even for a query whose search intention changes with time (time).

1 is a configuration diagram of a document search apparatus according to an embodiment of the present invention. The processing operation | movement figure of the same analysis processing stage. The processing data block diagram of the same analysis processing stage and an analysis processing stage. (A) is a figure which shows an example of a user's search action, (b) is a figure which shows an example of a binary graph. The figure which shows the processing operation of an analysis process stage and a query collation part. FIG. 6 is a data configuration diagram of search intention periodicity data, query search intention data, input query, and query input information. The figure which shows the processing operation of a search engine and a time interlocking type ranking process part. The data structure figure of the search result of a search engine, the final search result reranked, and click information.

  Hereinafter, a document search apparatus according to an embodiment of the present invention will be described. According to this document search device, a time-linked document search for grasping a user's search intention is performed according to the user's query input time (time). That is, information on which user has clicked the search log of the search engine is analyzed, and a series of search sections from the input of each user's query to the click of the search result and the end of the search are obtained.

  Next, paying attention to the first search action and the last search action in each section, grasp the difference and total number of different search intentions inherent in the query, and analyze the search intention in detail. The time change of the search intention of the analyzed query is analyzed using a long-term search log to find periodicity. A query in which periodicity is found means that a search intention of a user searched by the query changes with time (time).

  Therefore, if the same query is input again, the ranking of the search results is appropriately adjusted according to the input time and presented to the user. At this time, the information that the user really searched for at the time of inputting the query is presented in the upper rank of the search result, and the search accuracy is improved in this respect.

≪Example of device configuration≫
A configuration example of the document search apparatus will be described with reference to FIG. Here, the document search apparatus 1 is connected to a user terminal 2 so as to be able to transmit and receive data via a network, and captures a temporal change in search intention expressed by a query included in a search log for each user. A time-linked document search system that appropriately ranks search results according to the query input time (time) is configured.

  The user terminal 2 has a user interface 140 for recording queries, displaying search results, and a series of user search behaviors. The user interface 140 includes a query input unit 141 for inputting a query to be transmitted to the document search device 1, a search result display unit 142 for displaying a search result returned from the document search device 1, and a user search behavior. A search log recording unit 143 for monitoring and recording may be provided, and for example, it is configured by a browser of a mobile terminal such as a personal computer (PC), a mobile phone, or a PDA. The user's query input operation is performed through an input means such as a keyboard.

  The document search device 1 is constituted by a system server of a Web search engine (so-called search engine), and includes hardware resources of a normal computer such as a CPU, a memory (RAM), a hard disk drive device, a communication device and the like. As a result of the cooperation between the hardware resource and the software resource, the document search apparatus 1 analyzes the search intention of the query included in the search log of the search engine, and the query analyzed through the analysis process stage. An analysis processing stage for analyzing a temporal change in search intention of the user and a time-linked search ranking reflecting the temporal change in the search intention analyzed in the analysis processing stage in the search result are returned to the user terminal 2. And a search processing stage. The document search apparatus 1 does not necessarily have to be configured by a single computer, and may be configured by a plurality of computers.

  The analysis processing stage includes the search log DB 110. Query frequency totaling unit 112. Query frequency ranking DB 114. The analysis processing stage is performed through the search intention analysis unit 116, and the analysis processing stage includes a search intention time series analysis unit 120. This is implemented through the search intention periodicity DB 122. Both of these stages are performed in the offline stage before the query transmission from the user terminal 2. On the other hand, in the search processing stage, the query matching unit 130. Search engine 131 (including index DB 133). This is performed through the time-linked search result ranking processing unit 135. Of these, the DBs 110.114.122.133 are constructed in the hard disk drive. Data transmission / reception with the user terminal 2 is performed through the communication device.

  In brief, the analysis processing stage uses a search log recorded in the search log DB 110. Here, it is assumed that the search log DB 110 stores the record data of the search log recording unit 143 transmitted from the user terminal 2. The frequency counting unit 112 refers to the recorded data, and counts the appearance frequency of each query in the search log. The counted total results are transferred to the ranking DB 114 and accumulated. Next, the analysis unit 116 requests the ranking DB 114 for a high-frequency query set to be analyzed as a search intention. Upon receiving an answer to the request, the analysis unit 116 analyzes the search intention of each query belonging to the query set. Here, the search intention inherent in the query is extracted as a feature word (keyword) from the search behavior of the user, and the search intention time-series data in which the feature word is extracted in time series for each arbitrary period is transferred to the analysis unit 120.

  The analysis processing stage captures the temporal change of the query based on the search result time series data. Here, the analysis unit 120 analyzes the search intention time series data of the query transferred from the analysis unit 116, and if the feature word has periodicity, it is regarded as a keyword indicating periodicity. The analysis result, that is, the keyword indicating the period and the time (time) are registered in the periodicity DB 122 as search intention periodicity data, and are used for the ranking of the search results at the search processing stage.

  That is, the search processing stage generates a search ranking suitable for the query input time (time) from the user terminal 2 based on the registration data of the periodicity DB 122 and returns it to the user terminal 2. Here, when the user inputs / transmits a query to the query input unit 141, the document search apparatus 1 transfers the received query to the collation unit 130. At this time, the collation unit 130 accesses the periodicity DB 122 to collate the periodicity of the input query information. If the input query information has periodicity as a result of the collation, search intention periodicity data is acquired as search intention information of the query information.

  The acquired search intention information and input query information are transferred to the search engine 131. The search engine 131 collects Web page sets from the WWW (World Wide Web) 132 in advance and stores them in the index DB 133. Therefore, the search engine 131 performs a search based on the transferred input query information and obtains a normal search result. Is transferred to the ranking processing unit 135. At that time, search intention information is also transferred. Here, when the search intention information is transferred from the search engine 131, the ranking processing unit 135 reranks the search result based on the search intention information, and calculates the search result ranking that matches the query input time (time). Generate. The generated search result is returned to the user terminal 2 and displayed on the search result display unit 142.

  Accordingly, the user can browse the search result displayed on the search result display unit 142 and continue the search action such as clicking the search result. At this time, the search log recording unit 143 monitors a search action such as a user's query input or a click on a search result, and the information is transmitted to the document search device 1 and recorded / stored in the search log DB 110. In the following, the processing operation of each processing stage will be described in detail, roughly divided into an offline stage and an online stage.

≪Offline stage≫
(1) FIG. 2 shows an operation process related to an analysis process of a query retrieval intention in the analysis process stage. Here, it is assumed that the log 111 in FIG. 3 is stored in the search log DB 110. Specifically, the log 111 includes “input time / user ID / input query” at the time of query input by the user and “input time / user ID / input query / search result URL” at the time of clicking on the search result. Saved. This search result URL indicates the URL of the electronic document (Web page) clicked on the search result. Further, the user ID is a user identifier, and the user's search behavior, that is, a search session is acquired by analyzing the log 111 while fixing the user ID.

  The totaling unit 112 and the analysis unit 116 request a log to the search log DB 110 and acquire the log 111. Among these, the totaling unit 112 analyzes the acquired log 111 by paying attention to the number of input queries, and calculates the query frequency 113 (query frequency totaling step). In this query frequency 113, as shown in FIG. 3, it is assumed that the input frequency is stored for each query. The calculated query frequency 113 is transferred to the ranking DB 114 and stored.

  The analysis unit 116 requests a query set for performing a search intention analysis process to the ranking DB 114 together with a request for the log 111. Here, the analysis unit 116 acquires the processing target query 115 illustrated in FIG. 3, that is, a query set having a higher frequency with a large number of inputs. After acquiring the log 111 and the analysis target query 115, the search intention expressed by the query is analyzed (search intention analysis step).

  More specifically, the log 111 is analyzed with the user ID fixed for each query belonging to the processing target query 115, and the search behavior of each user is extracted. At that time, after entering the query, focus on the first search action (click on search result, change of query, or search end) and the last search action.

  The last search action means the search action immediately before the input of another keyword or the end of the search. Here, the first search action is referred to as “First Action”, and the last search action is referred to as “Last Action”. The reason for paying attention to “First Action” is that, for example, the user who first clicks on the first search result, such as the user of “UserID A” shown in FIG. 4A, and “UserID B” or “User ID C” This is because there is a user who skips the first search result and clicks on the second search result or the third search result first, like a user, and it is considered that the difference in the search intention for each user is easily apparent. is there.

  The reason why attention is paid to “Last Action” is that it can be considered that the search action satisfies the search intention and the search is terminated. Here, in order to clarify the difference in the search intention of each user expressed by the query and focus on the search behavior that was able to satisfy them, the connection between “First Action” and “Last Action” As a result, a binary graph shown in FIG. 4B is obtained.

  Then, the binary graph is analyzed, and a pair of “First Action” and “Last Action” having strong connection is calculated. Here, as an example, a method as shown in Non-Patent Document 2 (“3.3 Co-Visit Method (CVM), 3.4 Iterative Algorithm (IA))” and between each node (“First Action” and “ It is assumed that a similarity matrix of “Last Action”) is generated, and each user's search behavior is clustered by an existing clustering method such as an aggregation method. A feature word indicating a search intention inherent in the query from each cluster, for example, a query frequently used (clicked) to obtain a search result belonging to the cluster is assigned as a feature word. When using the method of Non-Patent Document 2, “Queries” and “Web Pages” are read as “First Action” and “Last Action”.

Such a search intention analysis is performed every week or every month, and as shown in FIG. 3, a search intention in which “date of query input, query, search intention number, feature word” is stored in pairs. Time series data 117 is acquired. Here, the query input time may be a season such as spring / summer / autumn / winter, or a time zone of morning / afternoon or morning / noon / night. A search intention number is assigned to each feature word corresponding to the input time. For example, the query 1 of the search intention time-series data 117 in FIG. 3 includes the phrase “tofu”, and the input periods are “summer” and “winter”. In this case, if the feature word of the input time “summer” is “cold” and the feature word of the input time “winter” is “yutofu”, different search intention numbers are assigned.
(2) Next, the processing operation of the analysis processing stage will be described with reference to FIG. 5. The analysis unit 120 performs analysis processing after transferring the search intention time-series data 117. Here, the presence or absence of periodicity of feature words in the search intention time-series data 117 is determined based on the search intention number through the fast Fourier transform method shown in Non-Patent Document 3 (search intention analysis step).

  If there is periodicity as a result of the analysis, a cluster feature word that typically appears in that cycle is regarded as a keyword characterizing the cycle. This analysis result is registered in the periodicity DB 122 as the search intention periodicity data 121. This processing operation ends the offline processing. In addition, as shown in FIG. 6, the search intention periodicity data 121 stores data indicating a period (season, time zone, etc.) for each query and feature words constituting the keyword in pairs.

≪Online stage≫
The processing operation in the document search stage in the search process stage will be described. First, the user inputs a query expressing the search intention to the query input unit 141. As shown in FIG. 6, the query 144 is composed of an input time (time) and a keyword set, and is transmitted from the user terminal 2 to the document search device 1. At this time, the query 144 is transferred to the search log recording unit 143 and stored in the data format of the query input information 145 to which the user ID information is added.

  As shown in FIG. 6, the query 144 is composed of an input time (time) and a keyword set, and is transmitted from the user terminal 2 to the document search device 1. At this time, the query 144 is transferred to the search log recording unit 143 and stored in the data format of the query input information 145 to which the user ID information is added.

  When the document search apparatus 1 receives the query 144, the document search apparatus 1 transfers the query 144 to the collation unit 130. The collation unit 130 accesses the periodicity DB 122, collates the query 144 with the query (query 1.2...) Of the search intention periodicity data 121, and confirms the periodicity (query collation step). More specifically, if there is a query that matches the query 144 in the search intention periodicity data 121, the periodicity is affirmed, and if there is no matching query, the periodicity is denied. Here, it is assumed that the matching of the keywords included in the query 144 is collated.

  If the periodicity of the query 144 is affirmed as a result of the collation, the search intention periodicity data 121 corresponding to the input time (time) is acquired as the search intention data 123 of the query. The search intention data 123 stores “query / time (time) / characteristic word”. However, when the periodicity is denied in the query 144 or when there is no search intention periodicity data 121 corresponding to the input time (time), empty data is configured as the search intention data 123. After the collation of the query, the collation unit 130 transfers the query 144 and the query intention data 123 of the query to the search engine 131.

  The search engine 131 collects a web page set (electronic document group) from a WWW (World Wide Web) 132 and stores it in the index DB 133 in advance. Therefore, as shown in FIG. 7, the search engine 131 searches the web page set based on the query 144 transferred from the collation unit 130 (search processing step), and thereby the search result 134 is acquired. The search result 134 and the query search intention data 123 acquired here are transferred to the ranking processing unit 135.

  The ranking processing unit 135 uses the transferred search results 134. A time-linked search result ranking is generated based on the search intention 123 (ranking processing step). That is, the search result 134 is reranked based on the query search intention 123. For example, among the search results 134, those including the characteristic words of the search intention data 123 are ranked higher. Accordingly, a search result ranking suitable for the search intention of the input query time (time) is generated, and the final search result 136 is returned to the user terminal 2 and displayed on the display unit 142. As a result, the user can confirm the final search result 136 and perform a search action such as a click on the browsed search result.

Click information 146 to which the click time and the user ID are assigned to the click information of the search result is transferred to the search log recording unit 143 and recorded. Of the search behavior of the user following the browsing of the final search result 136, a change (re-input) of the query is input through the query input unit 141, and the query input information 145 is also transferred to the search log recording unit 143 and recorded. Is done. At this time, the record data 145.146 of the search log recording unit 143 is transmitted to the document search apparatus 1, recorded in the search log DB 110, and fed back as the log 111 (recording step). Accordingly, the log 111 of the search log DB 110 is updated, and the updated search log DB 110 is used when generating a time-linked search result ranking of a query input from the same user next time. According to the document search apparatus 1 as described above, the following effects can be obtained.
(1) That is, according to the document search apparatus 1, since the log 111 of the search engine 131 is stored in the search log DB 110 with a fixed user ID, the query of each user is input through the analysis unit 116. A series of search sessions are acquired from the click to the click of the search result to the end of the search.

Here, paying attention to the first action (First Action) and the last action (Last Action) of each session and constructing a bipartite graph “Bipartite Graph” between them, “First Action” and “Last Action” Cluster the pair. These clusters represent different search intentions inherent in the query. By extracting feature words, the search intention can be grasped and the total number can be analyzed in detail, which contributes to improvement of search accuracy.
(2) Further, the analysis unit 116 converts the temporal change of the query analysis intention into time-series data based on a long-term search log, and discovers a change in periodicity through the analysis unit 120. A query in which a periodic change in search intention is found here means that the search intention of a user searched by the query changes with time (period). The search result ranking is appropriately reranked in accordance with the time (time) input in this manner and presented to the user.

  That is, the collation unit 130 confirms the periodicity of the input query based on the search intention periodicity data in the periodicity DB 122, obtains the appropriate search intention data 123 at the input time (time), and the ranking processing unit 135. A time-linked ranking is generated at.

  Therefore, the information that the user was really searching for at the query input time (time) is presented in the upper rank of the search result 136 displayed on the display unit 142. As a result, a highly accurate search result can be presented even for difficult queries whose search intent changes with time (time), and a time-linked document search device that takes time (time) into account is realized. This improves the search accuracy of the document search system.

≪Programs≫
The present invention can also be configured as a program that causes a computer to function as a part or all of each section 110.11.114.116.120.122.130.130.133.135 of the document search apparatus 1. According to this program, all or a part of each of the steps can be executed by a computer.

  This program can be provided through a network such as a website or e-mail. The program is recorded on a recording medium such as a CD-ROM, DVD-ROM, CD-R, CD-RW, DVD-R, DVD-RW, MO, HDD, Blu-ray Disk (registered trademark). It is also possible to save and distribute. This recording medium is read using a recording medium driving device, and the program code itself realizes the processing of the above embodiment, so that the recording medium also constitutes the present invention.

DESCRIPTION OF SYMBOLS 1 ... Document search device 2 ... User terminal 110 ... Search log DB
111 ... log 112 ... query frequency totaling unit (query frequency totaling means)
113 ... Query frequency 114 ... Query frequency ranking database 115 ... Processing target query (analysis target query)
116: Search intention analysis section (search intention analysis means)
117 ... Search intention time series data (search intention)
120 ... Search intention time series analysis unit (search intention analysis means)
121 ... Search intention periodicity data 122 ... Search intention periodicity DB (database)
123 ... Query search intention data 130 ... Query matching unit (query matching means)
131 ... Search engine 132 ... WWW (World Wide Web)
133 ... Index DB
134 ... Search results 135 ... Time-linked search result ranking processing unit (ranking processing means)
136 ... Final search result 140 ... User interface 141 ... Query input unit 142 ... Search result display unit 143 ... Search log recording unit 144 ... Query 145 ... Query input information 146 ... Click information

Claims (9)

A document search device that analyzes a user's search intention expressed by a query in a search log and reflects it in a search result of an electronic document group,
Query frequency counting means for counting the search frequency of each query for each user based on the number of input queries included in the search log;
A query set having a higher frequency in the search frequency counted by the query frequency counting means is analyzed, and pairs of the first search action and the last search action among the search actions for each user are clustered, and each cluster is represented by each cluster. Search intent analysis means for obtaining the search intent of the analyzed query,
A document search apparatus comprising: A search intent analysis means for determining the presence or absence of periodicity of the search intention extracted by the search intention analysis means;
2. The document search apparatus according to claim 1, wherein if the search intention has periodicity, it is reflected in the search result. The periodicity of the search intention of the user expressed by the query in the search log is determined, and the input time of the query input to the user terminal is used as the search result by using the database storing the search intention having the periodicity. A document retrieval device to be reflected,
The input query is checked against the stored data in the database to check the periodicity of the search intention of the input query, and if the periodicity can be confirmed, the stored data corresponding to the input time of the input query is acquired as the search intention data. Query matching means;
A ranking processing means for re-ranking a search result obtained by searching an electronic document group based on the input query based on search intention data acquired by the matching means;
A document search apparatus comprising: A document search device that searches for an electronic document group based on a query input to a user terminal and obtains a search result for responding to the user terminal,
A query frequency totaling unit that references a search log database that records a search log for each user and counts the search frequency of each query based on the number of times the query is included in the search log;
A query set having a higher frequency in the search frequency counted by the query frequency counting means is analyzed, and pairs of the first search action and the last search action among the search actions for each user are clustered, and each cluster is represented by each cluster. Search intention analysis means for acquiring search intention time-series data obtained by extracting the search intention of the analyzed query for each arbitrary period;
Search intention analysis means for determining periodicity in the search intention time-series data extracted by the search intention analysis means, and storing the periodicity in the periodicity database as search intention periodicity data if there is periodicity;
The input query is checked against the data stored in the periodicity database to check the periodicity of the search intention of the input query. If the periodicity can be confirmed, the search intention periodicity data corresponding to the input timing of the input query is searched. Query matching means to acquire as intention data;
Ranking processing means for reranking the search results based on the search intention data acquired by the query matching means,
A document search apparatus, wherein user click information for the input query and the final result is recorded in the search log database as needed. A document search method for analyzing a user's search intention expressed by a query in a search log and reflecting the result in a search result of an electronic document group,
A query frequency counting means for counting the search frequency of each query for each user based on the number of times the query is included in the search log;
The search intention analysis means analyzes the query set of the higher frequency in the search frequency counted in the query frequency aggregation step, clusters the pairs of the first search behavior and the last search behavior among the search behaviors for each user, A search intention analysis step for obtaining a search intention of an analysis target query represented by each cluster;
A document search method characterized by comprising: The search intention analysis means further includes a search intention analysis step for determining the presence or absence of periodicity of the search intention extracted by the search intention analysis step,
6. The document search method according to claim 5, wherein if the search intention has periodicity, it is reflected in the search result. The periodicity of the search intention of the user expressed by the query in the search log is determined, and the input time of the query input to the user terminal is used as the search result by using the database storing the search intention having the periodicity. A document search method to be reflected,
The query matching means checks the periodicity of the search intention of the input query by checking the input query against the stored data of the database. If the periodicity can be confirmed, the stored data corresponding to the input timing of the input query is searched. Query matching step to retrieve as intent data,
A ranking processing step in which the ranking processing means reranks the search results obtained by searching the electronic document group based on the input query based on the search intention data acquired by the matching means;
A document search method characterized by comprising: A document search method for searching for an electronic document group based on a query input to a user terminal and obtaining a search result for responding to the user terminal,
A query frequency counting unit that references a search log database that records a search log for each user and counts the search frequency of each query based on the number of times the query is included in the search log;
The search intention analysis means analyzes the query set of the higher frequency in the search frequency counted in the query frequency aggregation step, and clusters pairs of the first search behavior and the last search behavior among the search behaviors for each user, A search intention analysis step of acquiring search intention time-series data obtained by extracting the search intention of the analysis target query represented by each cluster every arbitrary period;
The search intention analysis means determines the periodicity in the search intention time-series data extracted in the search intention analysis step, and if there is a periodicity, the search intention analysis step stores the periodicity database as the search intention periodic data,
The query matching means checks the periodicity of the search intent of the input query by checking the input query against the data stored in the periodicity database. If the periodicity can be confirmed, the search intention according to the input time of the input query A query matching step for obtaining periodicity data as search intent data;
A ranking processing step, wherein the ranking processing means reranks the search result based on the search intention data acquired in the query matching step;
A recording step of recording user click information for the input query and the final result in the search log database as needed,
A document search method characterized by comprising:   A document search program for causing a computer to function as each of the means constituting the document search device according to any one of claims 1 to 4.

Images