JP2008027207A - Retrieval system and retrieval method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To enable to indicate retrieval result by a cluster easy to understand to a user. <P>SOLUTION: The retrieval system acquires a plurality of related words relating to a retrieval word from a retrieval query log (102), performs meta-search corresponding to the retrieval word by a plurality of retrieval engines (104), extracts text data from retrieved Web pages (108), acquires a plurality of vocabularies morphological-analyzing the extracted text data (112), makes a vocabulary frequency matrix to a plurality of Web pages (116), and calculates the degree of similarity of related words attending to only related words (120). Then, the system performs clustering of related words based on the degree of similarity of calculated related words, generates related word clusters of prescribed number (122), performs weighting of the related word clusters based on retrieval number of the related words, sorts the related word cluster systematically (124), and indicates a list of the related word clusters as retrieval result (128). <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a search device and a search method, and more particularly to a search device and a search method for searching document data by a search engine.

  Conventionally, various searches are performed using a Web search engine. When searching for information about popular items and phenomena, people, companies, products, services, TV programs, etc., it is not easy to narrow down the search results with appropriate related terms because you do not know much about the search target. There is a case.

  Further, even if the user knows the search target to some extent, there is a case where the user does not browse all the huge search results searched by the search word but wants an overview of only the page group that is interested.

In general, there are many similar documents included in the document set to be searched. Therefore, the document set is grouped (clustered) in advance according to the degree of similarity, and these groups (clusters) and A cluster type search for calculating the degree of matching with a search question (search query) is known (Non-Patent Document 1). The search result Web page group searched by a certain search term includes a large number of similar Web pages. Therefore, it is easy to narrow down the search result and overview the search result by performing appropriate clustering. Become.
Takenobu Tokunaga, “Information Retrieval and Language Processing”, The University of Tokyo Press, (1999)

  However, in the technique described in Non-Patent Document 1, when the search result Web page group is clustered with Web pages, the Web page group includes a lot of miscellaneous information that does not match the user's search needs. Clusters that do not make sense for the user and clusters that are not useful for narrowing down the search target are generated, so the clustered search results are difficult to understand for the user and less convenient. is there.

  The present invention has been made to solve the above-described problems, and an object of the present invention is to provide a search device and a search method that can display a search result using a cluster that is easy for the user to understand.

  In order to achieve the above object, a search device according to the present invention includes a document data acquisition unit that acquires a plurality of document data that matches a search term from a document database that stores a plurality of document data, and the document data acquisition unit. The frequency of calculating the appearance frequency of each of the plurality of related words related to the search word for each of the document data based on the word obtained by performing morphological analysis on each of the plurality of document data acquired by Calculation means; similarity calculation means for calculating the similarity between the related words based on the appearance frequency of each of the related words calculated by the frequency calculating means; and clustering of the related words. And generating a predetermined number of related word clusters by combining the related words from combinations having high similarity calculated by the similarity calculating unit. And clustering means, the related word clusters generated by the clustering unit, is configured to include a display means for displaying the search results matching the document data to the keyword.

  The search method according to the present invention acquires a plurality of document data that matches a search word from a document database storing a plurality of document data, and performs morphological analysis on each of the acquired document data. Based on the obtained word, for each of the document data, the appearance frequency of each of the plurality of related words related to the search word is calculated, and based on the calculated appearance frequency of each of the plurality of related words Calculating a similarity between the related words, clustering the plurality of related words, and generating a predetermined number of related word clusters by combining the related words from the combination having the calculated high similarity. The generated related word cluster is displayed as a search result of document data matching the search word.

  According to the present invention, a plurality of document data matching a search word is acquired from a document database storing a plurality of document data, each of the acquired plurality of document data is subjected to morphological analysis, and a word of the document data is obtained. obtain. And based on the obtained word, for each of the document data, to calculate the appearance frequency of each of a plurality of related words related to the search word, based on the calculated appearance frequency of each of the plurality of related words, The similarity between each related word is calculated.

  Then, a plurality of related words are clustered, and related words are combined from the calculated combination having a high degree of similarity to generate a predetermined number of related word clusters, and the generated related word clusters are adapted to the search word. Displayed as document data search results.

  Therefore, the search input by the user by displaying the result of clustering the related words based on the appearance frequency of each of the related words related to the search word in each document data matching the search word as the search result. Since related word clusters generated by excluding words irrelevant to words are displayed as search results, the search results can be displayed with clusters that are easy for the user to understand.

  Here, the related word related to the search word is a word that the user inputs to the search engine at the same time as the search word.

  Further, the search device according to the present invention provides, for each related word cluster generated by the clustering unit, a related word cluster among a plurality of document data acquired by the document data acquiring unit based on the appearance frequency of the related word. The display device further includes association means for associating the document data characterized by the related word with the related word cluster, and the display means displays the document data information indicating the related word cluster and the document data associated with the related word cluster as a search result. can do. As a result, the document data matching the search term is displayed in association with the related term cluster, so that the convenience of the user in displaying the search result can be improved.

  Further, the search device according to the present invention acquires a plurality of words as search terms simultaneously with the search terms in the document data acquisition unit based on a database storing a plurality of search queries consisting of at least one search term. The frequency calculation means can further calculate the appearance frequency of a plurality of related words acquired by the related word acquisition means for each piece of document data. Thereby, a plurality of related terms related to the search term can be acquired from the database storing the search query log.

  In addition, the search device according to the present invention provides a plurality of words as search terms simultaneously as synonyms of the search terms in the document data acquisition means based on a database storing a plurality of search queries consisting of at least one search term. It further includes related word acquisition means for acquiring, and the frequency calculation means can calculate the appearance frequency of a plurality of related words acquired by the related word acquisition means for each of the document data. Thereby, a plurality of related terms related to the synonym of the search term can be acquired from the database storing the log of the search query.

  Further, the display means according to the present invention can display the search results in order from the related word cluster including related words that are frequently searched simultaneously with the search word. Thereby, by displaying the related term cluster including the related terms that are strongly related to the search terms first, it is possible to meet the user's search needs and improve the convenience of the user in displaying the search results. it can.

  As described above, according to the search device and search method of the present invention, the result of clustering related words based on the appearance frequency of each of a plurality of related words related to the search word in each document data that matches the search word Is displayed as a search result, and related word clusters generated by excluding words irrelevant to the search word input by the user are displayed as the search result. Therefore, the search result is displayed in a cluster that is easy for the user to understand. The effect that it can be obtained.

  Hereinafter, the present embodiment will be described in detail with reference to the drawings. In the present embodiment, a case will be described in which the present invention is applied to a search device that collectively searches (metasearch) a plurality of search engines.

  As shown in FIG. 1, the search system 10 according to the first embodiment includes a search query log database 12 storing a search query log composed of a plurality of search queries, and a search query consisting of at least one search term. Corresponding to a plurality of Web search engines 14 that search Web pages, and a cache data database 16 that temporarily stores cache data of Web pages searched by the plurality of Web search engines 14, and a plurality of And a computer 18 storing a meta search engine program for realizing a meta search engine for performing a batch search (meta search) of the web search engine 14.

  The search query log database 12 stores a plurality of search queries including one or more search terms. Further, the plurality of Web search engines 14 use, for example, a plurality of main search engines (http://www.yahoo.co.jp/, http://search.msn.co.jp/) on the Internet. Yes.

  The computer 18 is connected to a morpheme analyzer 20 for morphological analysis of text data and a matrix calculation library 22 for performing matrix calculation.

  The meta search engine program is a program for executing a meta search processing routine, which will be described later, and a related word for acquiring a related word related to the input search word based on the search query log data acquired from the search query log database 12. A data acquisition module, a search data acquisition module that searches a Web page that matches a search term of a search query by a plurality of Web search engines 14 and temporarily stores cache data of the searched Web page in the cache data database 16; Matrix creation for the cache data of the cache data database 16 by performing morphological analysis by the morphological analyzer 20 to obtain words and creating a word frequency matrix indicating the appearance frequency of nouns and unknown words for each searched Web page Module, word A matrix generation module for performing matrix calculation on the degree matrix by the matrix calculation library 22, clustering related words to generate related word clusters, and a cluster ordering module for ordering the generated related word clusters Has been.

  The related term related to the search term is a word that is input by the user simultaneously with the search term as a search query to the search engine.

  Next, problems of conventional clustering of Web search results will be described. In the clustering method of existing documents, web pages, and web search results, when the entire document group or web page group to be clustered includes the user's search needs in a separable state, the user is interested. When only the document group possessed can be clearly separated from the entire document group, an effective cluster search can be performed.

  However, in general, the search result returned by the Web search engine is not an ideal Web page group for performing an effective cluster search, and often includes many miscellaneous information that is meaningless to the user.

  For example, consider a case where a word frequency matrix of a Web page as shown in FIG. 2 is created from the search result returned by the Web search engine for the search word “English conversation”. Web page clustering can be performed by calculating similarity in the Web page direction with respect to the word frequency matrix of the Web page, and word clustering can be performed by calculating similarity in the word direction. When all the matrix elements in FIG. 2 are used for clustering, words such as “case”, “month”, “day”, etc. that are not so much related from the viewpoint of the search term “English conversation” appear frequently or vice versa. If the appearance frequency is low and the rarity is high, or the feature amount of the word increases due to co-occurrence with other words, the clustering result is deteriorated. For example, according to the similarity calculation, when classification such as {English conversation learning, English diary, geos} {English conversation BBS, Eon} is performed in clustering of web pages, {case, day} is performed in word direction clustering. In both cases where the classification is {Month, Free, School, Teaching Material}, the clustering result is difficult for the user to understand.

  As described above, when the entire word frequency matrix constructed from the Web search results is used, miscellaneous information that does not match the user's search needs is affected. Therefore, in both cases of Web page clustering and word clustering, Cluster search cannot be performed effectively.

  Hereinafter, a meta search processing routine for displaying a clustering result of related terms as a search result based on a search term executed by the computer 18 will be described with reference to FIG.

  First, in step 100, it is determined whether or not the user has input a search word. When the user operates the keyboard or mouse (not shown) to input the search word, the process proceeds to step 102, where the search query log database 12 Then, related word data indicating a plurality of related words related to the search word is acquired.

  Here, Yahoo! recommends words that are useful for search. (R) or Yahoo! It is used for the related word search function of JAPAN (R) and keyword analysis of search advertisements. In search advertisements, information on related words input to a search engine by a user at the same time as a search word is used for keyword analysis. Is provided. For example, in the keyword analysis tool, it is information representing the search needs of a user who performs a search for the search word “English conversation”, and can be obtained with the search word “English conversation” by using related word data as shown in FIG. Search results can be narrowed down by information that matches the user's search needs.

  In the first embodiment, the related word data of the search word includes the keyword related tool of Overture (R) that can obtain the related word of the search word with an upper limit of 100 and the predicted value of the number of monthly searches. http://inventory.jp.overture.com/).

  In step 104, a meta search is performed using a plurality of web search engines 14 corresponding to the search term input in step 100, and a search result URL is sent from each of the web search engines 14 as web search result data. , Title, summary / snippet, and cache URL.

  Yahoo! (R), Yahoo! Major Web search engines such as JAPAN (R), Google (R), MSN Search (R), etc. prohibit access from unlicensed meta search engines and automatically access search sites provided for general users. Query transmission is prohibited, but instead, a search API and SDK for accessing search engine resources by a program are provided. For example, Google Web APIs (http://www.google.com/apis), Yahoo! Search Web Services SDK (http://developer.yahoo.net/search/), MSN Search Web Service SDK (http://msdn.microsoft.com/msn/)/msnsearch/! There is a JAPAN Web service SDK (http://developer.yahoo.co.jp/). In the first embodiment, Google Web APIs and Yahoo! A meta search is performed using the JAPAN Web service SDK.

  In the next step 106, cache data is acquired based on the cache URL, and the cache data is stored in the cache data database 16. In step 108, text data that is EUC-JP text is obtained from the HTML source file of the cache data. Extract.

  In step 110, the input search word and the related word acquired in step 102 are registered in the user dictionary of the morphological analyzer 20, and extracted in step 108 by the morphological analyzer 20 in step 112. The morphological analysis is performed on the text data to obtain a plurality of words as a morphological analysis result. In step 114, noise is removed from the morphological analysis result, and only nouns and unknown words around the search word are extracted. For morphological analysis, ChaSen (http://chasen.naist.jp/hiki/Chasen/) is used, and a search word or related word is registered in the ChaSen user dictionary, so that one word can be converted into a plurality of words. It is not divided into words.

  Then, in step 116, a word frequency matrix for a plurality of Web pages as shown in FIG. 2 is created from the extracted nouns and unknown words, and in step 118, a word that matches a related word in the created word frequency matrix. Column element IDs are extracted, and in step 120, the extracted column element IDs are specified, and the matrix calculation library 22 calculates the degree of similarity between related words focused only on related words.

  Here, as described above, when the similarity calculation in the word direction is performed on the entire word frequency matrix of the Web page created from the search results, miscellaneous information in the search results adversely affects clustering. End up. On the other hand, for example, the search results as shown in FIG. 2 are narrowed down by the related words included in FIG. 4, and the similarity is determined by the appearance frequency of only the related words “school”, “free”, and “learning material”. When calculated, as shown in FIG. 5, the words “case”, “month”, and “day” that are not important for the user who is interested in “English conversation” can be excluded from the similarity calculation targets.

  Then, in step 122, based on the similarity between the related words calculated in step 120, the related words are clustered, and the related words are combined from a combination having a high degree of similarity until a predetermined number of related word clusters are obtained. A combination of related words is performed from a combination having a high similarity, and a predetermined number of related word clusters are generated. For example, as shown in FIG. 5, by performing related word clustering limited to related words “school”, “free”, and “learning material”, a related word cluster such as {free, learning material} {school} is generated, As a page characterizing the related word cluster {free, learning material}, {English conversation learning, English diary, English conversation BBS} is associated with the related word cluster {free, learning material}, and as a page characterizing the related word cluster {school}, {Gios , Eon} is associated with the related word cluster {school}.

  In order to perform clustering of related terms, in the first embodiment, a general-purpose associative calculation engine GETA (http://geta.ex.nii.ac.jp/) is used as an associative calculation library. . In GETA, it is possible to specify a representative clustering distance calculation algorithm such as a single link method, a complete link method, a group average method, a WARD method, and hierarchical Bayesian clustering (HBC).

  In addition, related word clusters that match the search needs of many users can be generated by clustering related words limited by related words with a large number of searches (predicted value of the number of monthly searches).

In the next step 124, the related word clusters generated in step 122 are weighted based on the number of related word searches, the related word clusters are ordered based on the weight, and the related word clusters are sorted. The weight of the related word cluster C _i is calculated by the following formula.

ここで、ｆ_ｔは関連語クラスタＣ_ｉに含まれる関連語ｗ_ｔの検索数の総和であり、Ｔは関連語クラスタＣ_ｉに含まれる関連語の数である。

Here, f _t is the total number of searches of related words w _t included in the related word cluster C _i , and T is the number of related words included in the related word cluster C _i .

  For example, if the related word “children” is used in multiple searches such as “English conversation children” and “Children English conversation classes”, the total number of searches for the related word “children” is “English conversation children” “Children English conversation” The total number of searches for “classroom”. In the example of FIG. 4, the number of searches for “school” is 22,966, and the numbers of searches for “free” and “learning material” are 6647 and 2285, respectively. Accordingly, the weights of the related word cluster {free, learning material} {school} in FIG. 5 are calculated as 8932 and 22796, respectively.

  In step 126, the related word cluster is associated with the Web page indicated by the Web search data. In step 128, the sorted list of related word clusters is displayed as a search result as shown in FIG. Then, the meta search processing routine is terminated. When the related word cluster as shown in FIG. 5 is generated, the related word cluster is displayed in the order of {school} {free, learning material} in the search result. As described above, in the first embodiment, using related word data of search terms, related words are clustered using only related words frequently used in the search, and the generated related word clusters are further generated. Are weighted by the number of related word searches, and the related word clusters are sorted and the search results are displayed.

  In the display of the search result, as shown in FIG. 6, the title as the document data information of the Web page associated with the related word cluster is displayed as a detailed display of the related word cluster below the list of related word clusters. A summary and URL are also displayed.

  Next, a comparison experiment between the clustering of the first embodiment and the conventional clustering will be described. Here, as search terms, six examples of search words for cluster search exemplified on the top page of Clusty the Clustering Engine (http://clusty.jp/) (English conversation, nursing, mobile phone, stomach cancer, malicious business method) ), The results of related word clustering and web page clustering were compared.

  By changing conditions such as the number of related words used for clustering of related words, the number of clusters to be generated, and a distance calculation algorithm for clustering, the generated related word clusters are changed. Clustering of related terms and Web page clustering were performed under different conditions, and the clustering results were compared.

  FIG. 9 shows the results of clustering Web pages when related words are clustered under the conditions shown in FIG. 7, Web pages are clustered under the conditions shown in FIG. 8, and the search term is “English conversation”. In addition, as shown in FIG. 10, in the clustering result of the related words when the search word is “English conversation”, “free, teaching material, improvement method” indicates that the English conversation material is studied using the free English conversation material. "One-to-one, individual, private, lesson, instructor" can imagine if you want to take English conversation lessons personally, "Business, radio, daily life, travel" is a radio program You can imagine the case of listening and learning English conversation.

  FIG. 11 shows the clustering result of the Web page when the search word is “English conversation”, and FIG. 12 shows the clustering result of the related word when the search word is “English conversation”.

  In the above comparison results, the related word clustering result and the web page clustering result have almost no common point, and the web page clustering generates a cluster whose meaning is unrelated to the user's search intention. There was a tendency to be.

  On the other hand, in related word clustering, related words that seem familiar to users appear in the clustering results, and there is a tendency that related word clusters are generated for each user group and search purpose.

  Next, an evaluation experiment for comparing the result of clustering of related terms with the result of clustering of Web pages from the standpoint of a user who actually performs a search will be described. First, evaluators were 10 graduate students and university undergraduate students (male, early 20s), and the search words used were 20 words shown in FIG. Note that clustering was performed under the conditions shown in FIGS.

  The results of the related word clustering and the web page clustering were displayed side by side, and the evaluator asked the evaluator whether "which clustering result is easier to see" and obtained an answer. Ten evaluators compared the clustering results of 20 words, and a total of 200 responses were obtained. Out of 200 cases, 161 cases were “relevant word clustering results are easy to see” and 39 cases were “web page clustering results are easy to see”.

  Moreover, the answer results for each search term and each evaluator are shown in FIGS. 14 and 15, respectively. In addition, FIG. 16 shows the average browsing time per clustering result for each evaluator. Although the evaluation is divided according to the search term and the evaluator, it is presumed that the related word clustering can display a result that is easier to understand and view for the user than the Web page clustering.

  In the related word clustering in the first embodiment, synonyms (for example, “examination” and “simulation”), co-occurrence words (for example, “air ticket”, “vacant seat”, and “reservation”), sets (for example, "Lexus", "Harrier", "Isis" and "Wish"), notation shaking (for example, "Player" and "Player"), and compound words (for example, "Model" and "Change") each in one cluster There was a tendency to gather together. Based on this tendency, it can be said that the clustering of related words corresponds to the construction of a thesaurus of related words using a search result page group as a corpus.

  As described above, according to the search system according to the first embodiment, related words are clustered based on the appearance frequencies of a plurality of related words related to the search word in each Web page that matches the search word. The search results are displayed as search results, so that related word clusters generated by excluding words that are not related to the search terms entered by the user are displayed as search results. Can be displayed.

  Further, since the searched Web page is displayed in association with the related word cluster, the convenience of the user in displaying the search result can be improved.

  Also, a plurality of related terms related to the search term can be automatically acquired from the database storing the search query log.

  In addition, by displaying in order from related terms clusters that contain related terms that are frequently searched at the same time as the search terms, related terms clusters that contain related terms that are strongly related to the search terms are displayed first. It is possible to meet the needs, and it is possible to improve the convenience of the user in displaying the search results.

  In addition, the clustering of related terms using the related terms of the search terms frequently used by the user makes it possible to display a clustering result that is easy to understand for the user.

  Further, a large number of high-quality search results can be obtained by collectively searching a plurality of search engines.

  Further, by displaying a large number of obtained results in a clustered manner, it is possible to display a search result that is easy for the user to view.

  Further, by weighting the related word clusters by the number of related word searches, the related word clusters that are frequently referred to can be displayed at the top of the search results.

  In the above embodiment, the case where the computer obtains Web search result data using a plurality of existing search engines has been described as an example. However, the computer has a search engine function, and the Web A Web page that matches the search term may be acquired from a database that stores a plurality of pages. In this case, the related word acquisition and related word clustering functions are functions of the search engine.

  Moreover, although the case where a program such as a meta search processing routine is executed by a computer has been described as an example, the present invention is not limited to this, and the search system includes a portable information terminal. A program including a meta search processing routine may be executed.

  Next, a second embodiment will be described. In addition, about the component similar to 1st Embodiment, the same code | symbol is attached | subjected and description is abbreviate | omitted.

  The second embodiment is different from the first embodiment in that a search word for acquiring a related word and a search word for acquiring Web search data can be corrected and clustered again. ing.

  As shown in FIG. 17, the search system 210 according to the second embodiment includes a search query log database 12, a Web search engine 14, a cache data database 16, and a plurality of words in order to correct the search terms to synonyms. A computer 218 connected to a search term correction thesaurus database 212 storing synonyms for each of them is provided. Note that the synonyms include divided words and words that shake notation in addition to synonyms having a general meaning.

  In addition, a morphological analyzer 20 and a matrix calculation library 22 are connected to the computer 218.

  Next, a meta search processing routine according to the second embodiment will be described with reference to FIG. In addition, about the process similar to 1st Embodiment, the same code | symbol is attached | subjected and detailed description is abbreviate | omitted.

  First, in step 100, it is determined whether or not the user has input a search word. When the search word is input, in step 102, a relation indicating a plurality of related words related to the search word is displayed from the search query log database 12. Get word data. In step 104, a meta search is performed corresponding to the input search term, and Web search result data is acquired from each of the Web search engines 14, and in the next step 106, cache data is acquired and cache data is acquired. Store in database 16.

  In step 108, text data is extracted from the cache data. In step 110, the input search word and related word are registered in the user dictionary of the morphological analyzer 20. In step 112, the extracted text data is extracted. The morpheme analysis is performed to obtain a plurality of words as a morpheme analysis result. In step 114, noise is removed from the morpheme analysis result, and only nouns and unknown words around the search word are extracted.

  Then, in step 116, a word frequency matrix for a plurality of Web pages is created with the extracted nouns and unknown words, and in step 118, column element IDs of words that match related words in the created word frequency matrix are extracted. In step 120, the extracted column element ID is designated, and the similarity between the related words focused only on the related words is calculated.

  In step 122, related words are clustered to generate a predetermined number of related word clusters. In the next step 124, the generated related word clusters are weighted, and the related word clusters are ordered based on the weights. Sort related word clusters.

  In step 126, the related word cluster is associated with the Web page indicated by the Web search data, and in step 128, the sorted list of related word clusters is displayed as the search result.

  In the next step 230, it is determined whether or not the related word cluster displayed as the search result is to be corrected. If the user does not instruct the correction of the related word cluster, the meta search processing routine is terminated. When the related word cluster is instructed to be corrected by operating the keyboard or the mouse, it is determined in step 232 whether or not the related word data is to be corrected. If the user does not instruct the correction of the related word data, step 238 is performed. On the other hand, if the user operates the keyboard or mouse to instruct correction of related word data, the process proceeds to step 234.

  In step 234, a corrected search term is created for obtaining related term data. For example, a corrected search term is created by a user input, or a similar term of a search term is automatically acquired from the search term correction thesaurus database 212 to create a corrected search term. In the next step 236, related words searched simultaneously with the corrected search word are extracted from the search query log database 12 to obtain related word data, and the process proceeds to step 238.

  In step 238, it is determined whether or not the Web search result data is to be corrected. If the user does not instruct correction of the Web search result data, the process returns to step 106, and based on the newly acquired related term data, The related word cluster is generated again. On the other hand, if the user operates the keyboard or mouse to instruct correction of the Web search result data, the process proceeds to step 240.

  In step 240, a corrected search term is created for Web search result data acquisition. For example, a corrected search term is created by a user input, or a similar term of a search term is automatically acquired from the search term correction thesaurus database 212 to create a corrected search term. In the next step 242, a meta search is performed corresponding to the corrected search word, Web search result data is acquired from each of the Web search engines 14, the process returns to step 106, and the newly acquired related word data is acquired. Then, the related word cluster is generated again based on the Web search result data.

  As described above, according to the search system according to the second embodiment, after displaying a related word cluster as a search result, the search word is corrected and newly acquired related word data and Web search result data. Can be used to generate a related word cluster as a search result, so that the search result can be displayed in a cluster that is more easily understood by the user.

It is a block diagram which shows the search system which concerns on 1st Embodiment. It is an image figure which shows the word frequency matrix in a some web page. It is a flowchart which shows the content of the metasearch processing routine of the computer which concerns on 1st Embodiment. It is a table | surface which shows the number of searches with respect to the combination of a search word and a related word. It is an image figure which shows the word frequency matrix limited to the column element of a related word. It is an image figure of the search result display concerning a 1st embodiment. It is a table | surface which shows the conditions of clustering of a related word. It is a table | surface which shows the conditions of clustering of a web page. It is a figure which shows the clustering result of a web page when a search word is set to "English conversation." It is a figure which shows the clustering result of a related word when a search word is set to "English conversation." It is a figure which shows the clustering result of the web page at the time of setting a search term as "examination". It is a figure which shows the clustering result of a related word when a search term is set to "take an examination." It is a figure which shows the search term used for user evaluation. It is a graph which shows the legibility of the clustering result in each of a plurality of search terms. It is a graph which shows the visibility of the clustering result in each of several evaluators. It is a graph which shows the average browsing time of the clustering result in each of several evaluators. It is a block diagram which shows the search system which concerns on 2nd Embodiment. It is a flowchart which shows the content of the metasearch processing routine of the computer which concerns on 2nd Embodiment.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10,210 Search system 12 Search query log database 14 Search engine 16 Cache data database 18, 218 Computer 20 Morphological analyzer 22 Matrix calculation library 212 Thesaurus database for search term correction

Claims (6)

Document data acquisition means for acquiring a plurality of document data matching a search term from a document database storing a plurality of document data;
Based on words obtained by morphological analysis of each of a plurality of document data acquired by the document data acquisition means, for each of the document data, appearance of each of a plurality of related words related to the search word A frequency calculating means for calculating the frequency;
Similarity calculation means for calculating the similarity between each related word based on the appearance frequency of each of the plurality of related words calculated by the frequency calculation means;
Clustering means for generating a predetermined number of related word clusters by performing clustering of the plurality of related words and combining the related words from a combination having a high similarity calculated by the similarity calculating means;
Display means for displaying a related word cluster generated by the clustering means as a search result of document data matching the search word;
Search device including Each related word cluster generated by the clustering unit is characterized by a related word of the related word cluster among a plurality of document data acquired by the document data acquiring unit based on the appearance frequency of the related word. And further comprising an association means for associating the document data with the related word cluster,
The search device according to claim 1, wherein the display unit displays document data information indicating the related word cluster and document data associated with the related word cluster as the search result. Further comprising related word acquisition means for acquiring a plurality of words as search words simultaneously with the search words in the document data acquisition means based on a database storing a plurality of search queries consisting of at least one search word;
The search device according to claim 1, wherein the frequency calculation unit calculates an appearance frequency of a plurality of related words acquired by the related word acquisition unit for each of the document data. Further, related word acquisition means for acquiring a plurality of words as search words simultaneously with synonyms of the search word in the document data acquisition means based on a database storing a plurality of search queries consisting of at least one search word. Including
The search device according to claim 1, wherein the frequency calculation unit calculates an appearance frequency of a plurality of related words acquired by the related word acquisition unit for each of the document data.   The search device according to any one of claims 1 to 4, wherein the display unit displays the search results in order from related word clusters including related words that are frequently searched simultaneously with the search word. Obtain multiple document data that matches the search term from the document database that stores multiple document data,
Based on words obtained by morphological analysis of each of the plurality of obtained document data, for each of the document data, to calculate the appearance frequency of each of a plurality of related words related to the search word,
Based on the calculated appearance frequency of each of the related words, the similarity between the related words is calculated,
Performing clustering of the plurality of related words, combining the related words from the combination having the calculated high similarity, and generating a predetermined number of related word clusters;
A search method comprising: displaying the generated related word cluster as a search result of document data matching the search word.

Images