JP2002032394A - Related word information creation device, related word presentation device, document search device, related word information creation method, related word presentation method, document search method, and storage medium - Google Patents

Abstract

(57) [Summary] [Problem] In presenting related words, by presenting related words in clusters (groups) having a high degree of mutual relevance,
Help users select relevant words efficiently. A word clustering unit extracts a highly relevant subgraph as a cluster from a whole graph in which a word is a node and an association between words is an arc, and a node set of the obtained cluster is obtained. Then, the words in the registered documents in the document database unit 2 are clustered to create related word information and stored in the related word information storage unit 4. The related word information acquisition unit 6 refers to the related word information storage unit 4 and obtains related words of the word input by the word input unit 5 so that the related words can be presented for each cluster having a high degree of mutual relation.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a related word information creating apparatus and a related word information creating method for creating related word information to be referred to in order to obtain a related word of a given word, and a method for creating a word from the related word information. The present invention relates to a related word presentation device and a related word presentation method for finding a related word, a document search device and a document search method for performing a document search, and a storage medium storing a program for causing a computer to execute these processes.

[0002]

2. Description of the Related Art With the digitization of documents and the spread of the Internet, the need for document retrieval systems has increased.

In such a document search system, usually,
Search terms and natural language sentences ("" Natural language processing based support function in document search system Inforium "), by arranging search words and search character strings or combining them with AND / OR logical operators, Kameda et al. Proceedings of the 3rd Annual Conference pp357-350,1997.3 ”and“ CBSearch1000Ver.1.
3 '' homepage, http://www.justsystem.co.jp/produc
The user specifies a search sentence such as “t / system / cb13 / cb13.html, 1999”), and searches for a document that matches the search sentence from a document database to be searched.

[0004] Although the important words in the search sentence are the independent words, it is difficult to select an appropriate word, it is necessary to narrow down the search results, and it is necessary to add more words in order to prevent search omission. become. Therefore, the document database to be searched is analyzed in advance, the relevance between words is calculated, and when a search sentence is input, the related words of the word in the search sentence are presented, and the search sentence is input. There has been proposed a technology having a function of supporting a user who performs the operation (see Japanese Patent Application Laid-Open No. 10-334106). In addition, packaged software with this support function has been commercialized ("CBSearch10
00Ver.1.3 '' homepage, http: //www.justsystem.co.j
p / product / system / cb13 / cb13.html, 1999).

[0005]

However, in the related art for presenting related words, a list of words whose relevance is equal to or higher than the relevance using the related word of a word as a threshold or a list of words in order of relevance is provided. , In fact, even a single word,
There is "polysemy" used in various meanings and fields.

For example, as related words for the word “stadium”, J-League, Hanazono, Truck, Kashima, Goal, Try,
A list of related words such as athletics, ... is presented. These related words include words in three fields of rugby, soccer, and athletics, and are inconvenient for users.

Therefore, in the above example, if the related words of the "stadium" can be presented as follows: flower garden, goal, try,..., Kashima, J-League, goal,. Can assist the player in selecting related words efficiently (these correspond to the three fields of rugby, soccer, and athletics, respectively).

[0008] An object of the present invention is to provide a method for presenting related words in which related words are expressed in clusters (groups) having a high degree of mutual relation.
It is to support the user's efficient selection of the related word by presenting each related word.

An object of the present invention is to make it possible to present related words in clusters having a high degree of mutual relevance by using related word information created based on the degree of relevance between words. To be able to support the efficient selection of

[0010] An object of the present invention is to perform non-exclusive clustering so that related words can be presented for each cluster having a high degree of mutual relevance, thereby supporting efficient selection of related words by a user. It is to be.

An object of the present invention is to make it possible to present related words for each cluster having a high degree of mutual relevance by referring to a cluster dictionary and a word cluster dictionary, thereby enabling a user to efficiently select related words. I can help.

An object of the present invention is to make it possible to present related words for each cluster having a high degree of mutual relevance by using created related word information, and to assist a user in efficiently selecting related words. It is.

An object of the present invention is to enable only words having a high degree of relevance to a word for which a related word is to be obtained to be obtained as a related word.

It is an object of the present invention to output only a word having a high degree of relevance to a word whose related word is to be obtained as a related word.

It is an object of the present invention to output related words in the order of indices based on the degree of relevance to a word for which a related word is to be obtained.

An object of the present invention is to make it possible to present related words of a word included in a search sentence for each cluster having a high degree of mutual relevance, thereby preventing search omissions and improving the efficiency of search narrowing. It is.

[0017]

According to a first aspect of the present invention, a plurality of documents are stored in a related word information creating apparatus for creating related word information to be referred to in order to acquire a related word of a given word. A word clustering unit that clusters words included in the plurality of documents with reference to a document database unit that generates the related word information by classifying the words into clusters for each meaning or field. Is a related word information creating apparatus.

Therefore, it is possible to present related words for each cluster having a high degree of mutual relevance by using the created related word information, and it is possible to assist the user in efficiently selecting related words.

According to a second aspect of the present invention, in the related word information creating apparatus according to the first aspect, the word clustering unit performs the clustering based on a degree of association between the words.

Therefore, by using the related word information created based on the degree of relevance between words, it is possible to present related words for each cluster having a high degree of relevance to each other, and to provide the user with an efficient method of retrieving related words. We can help you make a choice.

According to a third aspect of the present invention, in the related word information creating apparatus according to the second aspect, the word clustering unit associates the words from the entire graph in which the words are nodes and the relationships between the words are arcs. The clustering is performed by extracting a subgraph having high possibility as a cluster and obtaining a node set of the obtained cluster.

Therefore, by performing non-exclusive clustering, it is possible to present related words for each cluster having a high degree of mutual relevance, thereby supporting the user to efficiently select related words.

According to a fourth aspect of the present invention, in the related-word information creating apparatus according to any one of the first to third aspects, the word clustering unit includes a cluster ID for identifying the cluster and a cluster ID belonging to the cluster. A word dictionary for associating the word ID identifying the word with the word dictionary and a word cluster dictionary associating the word ID of the word with the cluster ID of the cluster to which the word belongs for each word. It is characterized by being created as.

Therefore, referring to the cluster dictionary and the word cluster dictionary, it is possible to present related words for each cluster having a high degree of mutual relevance, thereby supporting the user to efficiently select related words. it can.

According to a fifth aspect of the present invention, there is provided a related-word information creating apparatus according to any one of the first to fourth aspects, and a related-word storing the related-word information created by the related-word information creating apparatus. An information storage unit, a word input unit that receives an input of a word for which a related word is to be obtained, a related word information acquisition unit that refers to the related word information storage unit and obtains a related word of a given word, And a related word information output unit for outputting related words.

Therefore, it is possible to present related words for each cluster having a high degree of mutual relevance by using the created related word information, and it is possible to assist the user in efficiently selecting related words.

According to a sixth aspect of the present invention, in the related word presenting apparatus according to the fifth aspect, the related word information acquiring section includes:
It is characterized in that, among words that can be obtained as the related words by referring to the related word information storage unit, only those having a high degree of relevance to the word input by the word input unit are obtained as the related words.

Therefore, only words having a high degree of relevance to the word for which a related word is to be obtained can be obtained as related words.

According to a seventh aspect of the present invention, in the related word presenting apparatus according to the fifth aspect, the related word information output unit includes:
The related words obtained by the related word information acquisition unit are limited to those having a high degree of relevance to the word input by the word input unit, and are output.

Therefore, only words having a high degree of relevance to the word for which a related word is to be obtained can be output as related words.

According to an eighth aspect of the present invention, in the related word presenting apparatus according to any one of the fifth to seventh aspects, the related word information output unit includes the related word present in the same cluster. The output is performed in the order of indices based on the degree of relevance to the word input by the word input unit.

Therefore, related words can be output in the order of indices based on the degree of relevance to the word whose related word is to be obtained.

According to a ninth aspect of the present invention, a search sentence input unit for receiving a search sentence input, and a document to be matched with the input search sentence are searched by referring to a document database unit storing a plurality of documents. A search unit, a search result output unit that outputs the obtained search result, and a word input unit,
9. The apparatus according to claim 5, further comprising a word acquisition unit that extracts a word from the search sentence input by the search sentence input unit and inputs the extracted word to the related word information acquisition unit. 9. And a search sentence correction unit that corrects the search sentence input by the search sentence input unit based on the related words obtained by the related word presentation device based on all or specified words. And a document search device.

Therefore, it is possible to present related words of words included in a search sentence for each cluster having a high degree of mutual relevance, thereby preventing search omissions and improving the efficiency of search narrowing.

According to a tenth aspect of the present invention, in a related word information creating method for creating related word information to be referred to to obtain a related word of a given word, a reference is made to a document database section in which a plurality of documents are registered. A word clustering step of clustering words included in the plurality of documents, classifying the words into clusters for each meaning or field, and creating the related word information. This is a word information creation method.

Accordingly, it is possible to present related words for each cluster having a high degree of mutual relevance by using the created related word information, and it is possible to assist the user in efficiently selecting related words.

According to an eleventh aspect of the present invention, in the related word information creating method according to the tenth aspect, the word clustering step performs the clustering based on a degree of association between the words.

Therefore, it is possible to present related words for each cluster having a high degree of mutual relevance by using related word information created based on the degree of relevance between words, and to provide a user with an efficient method of relevance of related words. We can help you make a choice.

According to a twelfth aspect of the present invention, in the word clustering step, a complete graph or a transitivity graph slightly loosened from the whole graph in which the word is a node and the relationship between the words is an arc is defined as a cluster. 12. The related-word information creating method according to claim 11, wherein the clustering is performed by extracting and obtaining a node set of the obtained cluster.

Therefore, non-exclusive clustering is performed, and related words can be presented for each cluster having a high degree of mutual relevance, so that the user can be efficiently selected.

The thirteenth aspect of the present invention relates to the tenth aspect.
12. In the related-word information creating method according to any one of the first to twelfth, the word clustering step includes: a cluster dictionary in which a cluster ID for identifying the cluster is associated with a word ID for identifying the word belonging to the cluster; For each word, a word cluster dictionary in which the word ID of the word and the cluster ID of the cluster to which the word belongs is created as the related word information.

Therefore, referring to the cluster dictionary and the word cluster dictionary, it is possible to present related words for each cluster having a high degree of mutual relevance, thereby assisting the user in efficiently selecting related words. it can.

According to a fourteenth aspect of the present invention, there is provided a word input step for receiving an input of a word whose related word is to be obtained,
A related word information storage unit that stores the related word information created by the related word information creating method according to any one of 0 to 13; , And a related word information output step of outputting the obtained related word.

Accordingly, it is possible to present related words for each cluster having a high degree of mutual relevance by using the created related word information, and it is possible to assist the user in efficiently selecting related words.

According to a fifteenth aspect of the present invention, in the related word presentation method according to the fourteenth aspect, the related word information obtaining step includes the step of obtaining the related word by referring to the related word information storage unit. The word related to the word input by the word input unit is limited to those having a high degree of relevance, and is determined as the related word.

Therefore, only words having a high degree of relevance to the word for which a related word is to be obtained can be obtained as related words.

According to a sixteenth aspect of the present invention, in the related word presentation method according to the fourteenth aspect, the related word information output step includes the step of outputting the related word information of the related word obtained by the related word information acquisition unit. And output only those having a high degree of association with the input word.

Therefore, only words having a high degree of relevance to the word for which a related word is to be obtained can be output as related words.

According to the seventeenth aspect of the present invention,
16. The related-word presentation method according to any one of 16.
In the related word information output step, the related words in the same cluster are output in the order of indices based on the degree of relevance to the word input by the word input unit.

Therefore, related words can be output in the order of indices based on the degree of relevance to the word whose related word is to be obtained.

According to an eighteenth aspect of the present invention, there is provided a search sentence input step of receiving a search sentence input, and searching for a document to be collated with the input search sentence by referring to a document database section storing a plurality of documents. A search step, a search result output step of outputting the obtained search result, and a word extracted from the search sentence input in the search sentence input unit instead of the word input step. And a word obtaining step of providing the related word information obtaining step.
17. The related-word presenting method according to any one of No. 17, wherein the related-word presenting step of obtaining the related word based on the all or specified words, and the search sentence input step based on the obtained related word And a search sentence correcting step of correcting the input search sentence.

Therefore, it is possible to present related words of words included in a search sentence for each cluster having a high degree of mutual relevance, thereby preventing search omissions and improving the efficiency of search narrowing.

According to the nineteenth aspect of the present invention, a computer storing a related word information creating program for causing a computer to create related word information to be referred to to acquire a related word of a given word can be read by a computer. In a simple storage medium, the related word information creating program clusters words included in the plurality of documents by referring to a document database unit in which the plurality of documents are registered, and clusters the words in terms of the meaning or the cluster for each field. And causing the computer to execute a word clustering step of classifying the related term information and creating the related term information.

Accordingly, it is possible to present related words for each cluster having a high degree of mutual relevance by using the created related word information, and it is possible to assist the user in efficiently selecting related words.

According to a twentieth aspect of the present invention, in the storage medium according to the nineteenth aspect, in the word clustering step, the clustering is performed based on a degree of association between the words.

Therefore, by using the related word information created based on the degree of relevance between words, it is possible to present related words in clusters having a high degree of relevance to each other, so that the user can efficiently recognize related words. We can help you make a choice.

According to a twenty-first aspect of the present invention, in the storage medium according to the twentieth aspect, in the word clustering step, a complete graph or a complete graph is obtained from the entire graph in which the word is a node and an association between words is an arc. It is characterized in that the transitivity graph in which is slightly relaxed is extracted as a cluster, and the clustering is performed by obtaining a node set of the obtained cluster.

Therefore, by performing non-exclusive clustering, it is possible to present related words for each cluster having a high degree of mutual relevance, and it is possible to assist the user in efficiently selecting related words.

The invention described in claim 22 is the invention according to claims 19 to
21. The storage medium according to any one of 21., wherein the word clustering step includes, for each word, a cluster dictionary that associates a cluster ID for identifying the cluster with a word ID for identifying the word belonging to the cluster. A word cluster dictionary in which the word ID of the word and the cluster ID of the cluster to which the word belongs is created as the related word information.

Therefore, referring to the cluster dictionary and the word cluster dictionary, it is possible to present related words for each cluster having a high degree of mutual relevance, thereby assisting the user in efficiently selecting related words. it can.

According to the twenty-third aspect of the present invention,
22. The storage medium according to any one of 22., wherein a word input step of receiving an input of a word for which a related word is to be acquired, and the related word information created by a computer based on the related word information creation program are stored. A related word information obtaining step for obtaining a related word of the word given in the word input step with reference to a related word information storage unit, and a related word information outputting step of outputting the obtained related word, to a computer A related word presentation program to be executed is stored.

Therefore, it is possible to present related words for each cluster having a high degree of mutual relevance by using the created related word information, and it is possible to assist the user in efficiently selecting related words.

According to a twenty-fourth aspect of the present invention, in the storage medium according to the twenty-third aspect, the related-word information obtaining step includes:
It is characterized in that, among words that can be obtained as the related words by referring to the related word information storage unit, only those having a high degree of relevance to the word input by the word input unit are obtained as the related words.

Therefore, only words having a high degree of relevance to the word for which a related word is to be obtained can be obtained as related words.

According to a twenty-fifth aspect of the present invention, in the storage medium according to the twenty-third aspect, the related-word information output step includes:
The related words obtained by the related word information acquisition unit are limited to those having a high degree of relevance to the word input by the word input unit, and are output.

Therefore, only words having a high degree of relevance to the word for which a related word is to be obtained can be output as related words.

The invention according to claim 26 is the invention according to claims 23 to
25. The storage medium according to any one of 25, wherein the related word information output step is such that the related words in the same cluster are arranged in the order of indices based on the degree of relevance with the word input in the word input unit. It is characterized by outputting.

Therefore, related words can be output in the order of indices based on the degree of relevance to the word whose related word is to be obtained.

The invention according to claim 27 is the invention according to claims 23 to
26. The storage medium according to any one of 26, wherein a search sentence input step of receiving an input of a search sentence and a document to be matched with the input search sentence are referred to by referring to a document database unit storing a plurality of documents. A search step for searching, a search result output step for outputting the obtained search result, and a word extracted from the search sentence input in the search sentence input unit instead of the word input step, and the extracted words are extracted. A related word presenting step of providing the related word information to the related word information acquiring step, and a related word presenting program for causing a computer to execute the related word presenting step of finding the related word based on all or specified words; A document search program for causing a computer to execute a search sentence correcting step of correcting the search sentence input in the search sentence input step based on a word.

Accordingly, it is possible to present related words of words included in a search sentence for each cluster having a high degree of mutual relevance, thereby preventing search omissions and improving the efficiency of search narrowing.

[0071]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS [First Embodiment of the Invention] One embodiment of the present invention will be described as a first embodiment of the invention.

FIG. 1 is a functional block diagram of a related word presentation device 1 according to the first embodiment of the present invention. As shown in FIG. 1, the related word presentation device 1 includes a document database unit 2. This is a document database for registering and managing a plurality of documents, and is managed using, for example, a relational database management system. The word clustering unit 3 extracts words from each document registered in the document database unit 2, checks the relation between the words, and clusters the words based on the extracted words, and implements a related word information creating device. is there. The related word information storage unit 4
The result of word clustering by the word clustering unit 3 is stored as related word information. The word input unit 5 receives an input of a word to obtain a related word, and sends the input to the related word information acquiring unit 6. The related word information acquisition unit 6 searches the related word information storage unit 4 for the input word sent from the word input unit 5, acquires related word information, and outputs the result to the related word information output unit 7. The related word information output unit 7 outputs the related word information sent from the related word information acquisition unit 6.

FIG. 2 is a block diagram showing a hardware configuration of the related word presentation device 1. As shown in FIG. 2, the related word presentation device 1 includes a CPU 11, a ROM 12, a RAM
13 are connected by a bus line 14. The bus line 14 is provided with an LC through a predetermined interface.
D, an output device 15 such as a printer, an input device 16 such as a keyboard and a pointing device, and a hard disk 17. Further, a storage medium reading device 19 for reading the storage medium 18 and a communication control device 2 for connecting the related word presentation device 1 to a network such as the Internet.
0 and at least one of them.

A related word presentation program is installed on the hard disk 17. This related word presentation program includes a related word information creation program.
The related word presentation program is stored in the storage medium 18 and can be read by the storage medium reading device 19 and installed on the hard disk 17. Further, a related word presentation program can be downloaded from a network such as the Internet and installed on the hard disk 17.

The storage medium 18 is, specifically, a CD, a DV,
Various types of media such as various types of optical disks such as D, various types of magneto-optical disks, and various types of magnetic disks such as FDs can be used. Specifically, the storage medium reading device 19 includes a CD drive device, a DVD drive device, and a F drive, depending on what medium is used as the storage medium 18.
D drive devices, MO drive devices, and the like are used.
A storage device that stores a related-word presentation program in a server on the transmission side when downloading from a network such as the Internet and installing it on the hard disk 17 is also a storage medium of the present invention.

The document database unit 2 has a hard disk 1
7 is built. By installing the related word presentation program on the hard disk 17, the CPU 11
Executes the functions of the word clustering unit 3, the related word information storage unit 4, and the related word information acquisition unit 6, based on the related word presentation program. The storage area of the related word information storage unit 4 is secured in the RAM 13 or the hard disk 17. Further, the function of the word input unit 5 is executed via the input device 16, and the function of the related word information output unit 7 is executed via the output device 15. The related word presentation program may operate on a predetermined OS. Further, a part of the function may be performed on behalf of the OS or other application software.

Next, the processing performed by the related word presentation device 1 based on the related word presentation program will be described. FIG.
4 is a flowchart illustrating a process performed by the related word presentation device 1.

(A) Creation / Storage of Related Word Information The purpose of the related word presentation device 1 is to “output related words to the related word information output unit 7”. When the unit 2 is updated, the processing of “creation / storage of related word information” is executed. FIG. 3 is a flowchart of the process of “creating / storing related word information” performed by the related word presentation device 1. With this processing, a related word information creating method is realized.

[0079] 3. Word Extraction from Documents As shown in FIG. 3, first, in step 1, each document registered in the document database unit 2 is subjected to word segmentation with part of speech by a morphological analysis process to extract words (step 1). S1).

The words to be extracted may cover a wide range of independent words including adjectives and verbs, but when the related word presentation device 1 is used for a document search system which is the most common application. Often, only general nouns are excluded, excluding functional nouns such as formal nouns and adverbial nouns. Here, a description will be given of only this general noun. Further, compound nouns can be decomposed into more basic nouns, and there is a problem of what unit is to be taken. In this case, compound nouns are extracted as they are without being decomposed.

For example, the following document 1 (document ID = 1 /
By performing morphological analysis on each sentence in the document and extracting nouns from the Asahi Shimbun (an article on January 21, 1993), words (nouns) appearing in document 1 as shown in Listing 1 below Is obtained, and the following list 2 is obtained by eliminating duplication of the same word.

<Document 1> ------------------------------------------- ------------------- Export control of industrial products related to conventional weapons has started. Export control for four countries. It began in Japan and Japan. This is a regulation based on the agreement of the seven major advanced nations (G7), including the United Kingdom and the United States, and applies the list of the Communist Border Export Control Committee (COCOM) mutatis mutandis to four countries, including Iraq. The G7 has already begun discussions to expand its target items and nations and is likely to develop new export controls that will change at the end of the Cold War.

The four regulated countries are Iran, Iraq, Libya, and the Democratic Republic of Korea (North Korea). North Korea has already been designated as a communist bloc, so three new countries will be added. Iraq is also embargoed on economic sanctions. Major changes in exports under this measure are likely to be in Iran and Libya.

When exporting a regulated item such as a computer or a machine tool set based on an export trade control order or the like, a license application is made to the Ministry of International Trade and Industry. At that time, permission will not be granted to countries designated as specific areas unless they are clearly for consumer use, and in fact restricted products cannot be exported. -------------------------------------------------- ------------

<List 1> ------------------------------------------- ------------------- Weapons Export Control Conventional Weapons Industrial Products Export Controls Targets Conventional Weapons Parts Processing Machinery Industrial Products Export Controls Japan UK United States Major Advanced G7 Agreement Regulation Iraq Target Anti-Communist Area Export Control Committee Cocom List G7 Subject Items Discuss Cold War End Export Control Restrictions Iran Iraq Libya North Korea Democratic Republic of the Republic of North Korea Computer Machine tools Controlled items Ministry of International Trade and Industry Permit application Specific area Consumer license Restricted products ---------------------------------- ---------------------------

<List 2> ------------------------------------------- ------------------ Weapons Export Regulations Conventional Weapons Industrial Products Target Parts Processing Machinery Japan UK United States Major Advanced G7 Agreement Regulations Iraq Export Control Committee for Communist Areas Cocom List Items Discuss End of the Cold War Iran Libya DPR Korea North Korea Communist Specified Area Economic Sanctions Embargo Measures Export Changes Export Trade Control Ordinance Computer Machine Tools Controlled Items Ministry of International Trade and Industry Permit application -------------------------------------------------- ---

Further, by arranging words appearing for each document, a word list for each document as shown in Table 1 is obtained.

[0088]

[Table 1]

By re-creating the word in Table 1 as a key, Table 2
An index with a word ID as a key can be created. This index is similar to a search index prepared by a document search system that uses words as search keywords. That is, a set of words and a document ID is generated, and the set data is sorted using a word notation as a key.
The D list can be obtained, and the word ID can be easily obtained by sequentially assigning different words to different words.

Since the words are sorted using the word display as a key, the word IDs are in a so-called dictionary order.

[0091]

[Table 2]

Further, using the obtained word ID, Table 1
By replacing the words in the word list for each of the document IDs with the word IDs and sorting by the word ID value, an index with the document ID as a key as shown in Table 3 can be created.

[0093]

[Table 3]

[0094] Word Clustering In the related art, in the related art, a relevance (similarity) between words is defined, and a word having a high relevance is selected as a related word. As an index used as the degree of association, mutual information (“Word association norms, mutual information,
and lexicography ”, Kenneth Ward Church and patric
Hanks, Computational Lingusitics, 16 (1): 22-29, 1990
), Yule Y, DiceCofficient, etc.

In the first embodiment of the present invention, it is considered that related words of a certain word belong to different clusters (groups) according to meanings and fields, and the related words are presented by organizing each cluster. Because of this, word clustering is performed after step S1 (step S2). This implements a word clustering step.

For this purpose, when automatic processing is performed by information processing instead of manual creation, it is common to use the degree of association between words. However, in the clustering method of merging the most relevant ones, which is most commonly considered, only an exclusive cluster is generated, and therefore, a certain word is not presented for each of a plurality of clusters. As one of the clustering processing methods that allow duplication, "clustering of co-occurrence graph based on transitivity graph"("clustering of co-occurrence graph based on transition rate",
Kumiko Tanaka et al., IPSJ Technical Committee Report Natural Language Processing 115
-15,1996 and “Clustering Co-occurence Graph
based on Transitivity ”, Kumiko TANAKA-Ishii, Hideya
IWASAKI, The 5th Workshop on Very Large Corpora (WV
LC5), 1998). In this word clustering, a highly relevant graph is extracted as a cluster from the entire graph in which a word is a node and an association between words is an arc, and a node set of the obtained cluster is used as a related word cluster. . Specifically, a complete graph in which words are nodes and the relationships between words are arcs, and a transitive rule graph that is slightly relaxed are extracted as clusters, and the node set of the obtained clusters can be used as related word clusters. Conceivable.

(1) Mutual Information Amount of Relevance Hereinafter, here, word clustering is performed as follows:
A process of performing “clustering of co-occurrence graph based on transitive rule graph” using the mutual information amount that is best known as the degree of association will be described, but the present invention is not limited to this means.

The following equation shows the word W in N documents.
This is an expression that defines the mutual information amount M (Wa, Wb) between a and the word Wb.

M (Wa, Wb) = log (P (Wa, Wb)
/ P (Wa) P (Wb)) where P (Wn) = Nn / N, P (Wa, Wb) =
Nab / NP (Wn) is the probability that the word Wn appears in N documents, and Nn is the number of documents in which the word Wn appears. P (Wa, Wb) is the probability that the word Wa and the word Wb co-occur in N documents, and Nab is the number of co-appearing documents. The unit of co-occurrence may be a document part or a plurality of documents. However, here, an example in which the document is a unit will be described.

The number of documents and the number of appearing documents for each word are obtained in step S1. For the mutual information amount, the number of documents in which two words co-appear is required for every word pair. In order to obtain this value, for example, as the simplest means, the document I of the appearing document of each of the word and the word larger than the word ID is used in ascending order of the word ID.
What is necessary is to match the list of D and count the number of matches.

For example, as shown in Table 2, "G7" having a word ID of 3 is sequentially searched from the word having a word ID of 4. For example, the number of co-occurring documents of "America" having a word ID of 13 is as follows. The list of document IDs of the number of appearing documents 6 of “G7” in Table 2 is “1,88,105,288,300,410”, and the list of document IDs of the number of appearing documents of “America” of 42 is “2,3,5,88,1”.
05,109,300,310,479,… ”
5,300 "in common.

Table 4 shows, among the matrices indicating the number of co-occurring documents between all the appearing words thus obtained (with no lower left corner from the diagonal portion), “G7” (word ID = 3), “J-League” (4), only the part of the number of co-occurring documents of 10 words is extracted. Note that “Nn” in Table 4 is the “number of appearing documents” of the word, and a value is shown in the corresponding column.

[0103]

[Table 4]

From Table 4, for example, for the word “G7” and the word “America”, the number of single appearing documents Nn is 6,42 and the number of co-occurring documents is 3, and the mutual information amount of these two words is obtained. Can be obtained as follows.

P (“G7”) = 6/2000 = 0.003 P (“America”) = 42/2000 = 0.021 P (“G7”, “America”) = 3/2000 = 0.015 M (“G7”, “G7” America ") = log (P (" G7 "," USA ") / P (" G7 ") P (" America ") = log (3 ・ 2000 / (6.42)) = log (1000/42) ≒ 4.57 (Here, the basis of log is set to 2.) The above calculation is performed for all the combinations of words, and the mutual information between all the words can be obtained. This is a table in which the mutual information amount is extracted.

[0106]

[Table 5]

The mutual information is considered to indicate the degree of relevance between two words. For example, if a threshold value of 3 is used, in Table 5, among the combinations of words whose mutual information exceeds 0, “US (13) and "goal" (37), "try" (65) and "Kashima" (502) are regarded as having no relationship between words because the relevance is below the threshold.

(2) Clustering of co-occurrence graph based on transitivity graph Extraction of cluster Gi (set of node set and arc set) from whole graph G (set of all node set and all arc set) by this method The procedure is as follows ("" Clustering Co-occurrence Graph Based on Transition Rate ", Kumiko Tanaka et al.,
IPSJ SIG Technical Report 115-15, 1996 "and""Clustering Co-occurence Graphbased on Trans
itivity ", Kumiko TANAKA-Ishii, Hideya IWASAKI, The
5th Workshop on Very Large Corpora (WVLC5), 1998 ”).

[0] Assuming that i = 0, a cluster Gi (G0) in which both the node set and the arc set are empty is created.

[1] If there is an arc e which is not included in any of G0,..., Gi and a node v which forms a triangle with arc e, all nodes and arcs forming this triangle are put into cluster Gi + 1. If there is no graph (node and arc) to be put into the cluster Gi + 1, the processing ends.

[2] The arc e included in the cluster Gi + 1 forms a triangle with the node v not included in the cluster Gi + 1. If there is an arc e not included in any of G0,... Node v is Gi
If it is connected to the node v 'included in +1, all arcs of this triangle and the arc between the node v and the node in Gi + 1 are put into the cluster Gi + 1. If the cluster Gi + 1 cannot be expanded, the process proceeds to [4].

[3] Return to [2].

[4] Add 1 to i and return to [1].

Although the example shown above consists of a large number of words, for the sake of explanation, the process of the above cluster extraction procedure will be described by using a graph having the words specified in Table 5 as nodes as an entire graph. Is shown.

Assuming that the mutual information threshold is 3, the nodes and arcs are as follows. -------------------------------------------------- ----------- ・ Nodes: (3), (4), (13), (37), (65), (66), (184), (414), (50 2) , (974) ・ Arc: (3)-(37), (4)-(37), (4)-(414), (4)-(502), (37)-(65), (37) -(184), (37)-(414), (37)-(502), (65)-(184), (65)-(414), (66)-(414), (66)-( 974), (184)-(414), (414)-(502), (414)-(974) ------------------------ -------------------------------------

FIG. 4 is a graph showing the result.

According to the procedure described above, the cluster extraction processing proceeds in the following manner.

[0] With i = 0, an empty cluster Gi is created for both the node set and the cluster set.

[1: Initial Setting of G1] The arc (3)-(37) not included in G0 has no node forming a triangle with the arc (3)-(37). If you select the next arc (4)-(37), this arc and node (414) will be arc (4)-(414), arc (37)
-(414) forms a triangle, so G1 is <
{(4), (37), (414)}, {(4)-(37), (4)-(41
4), (37)-(414)}> is initialized.

[2: Addition of G1] Arcs (4)-(37) included in G1 are nodes (502) not included in G1, arcs (4)-(502), and arcs (37)-(502). ) Form a triangle, and are connected by nodes (414) included in G1 and arcs (414)-(502). From the above, G1
, A node (502), arcs (4)-(502), arcs (37)-(502), and arcs (414)-(502) are added.

[3] Return to [2].

[2: Addition of G1] Since any arc included in G1 does not form a triangle with a node not included in G1, the processing of G1 ends.

At this point, the following is performed. -------------------------------------------------- ----------- G0 = <{}, {}> G1 = <{(4), (37), (414), (502)}, {(4)-(37), (4)-(414), (4)-(502), (37)-(414), (37)-(502), (414)-(502)}> Arc not belonging to any class: (3)-(37), (37)-(65), (37)-(184), (65)-(184), (65)-(414), (66)-(414), (66 )-(974), (184)-(414), (414)-(974) ---------------------------- ---------------------------------

[4] It is assumed that i = 1.

[1: Initial Setting of G2] Arcs (37)-(65) not included in G0 and G1 are nodes (414) having arcs (37)-(414) and arcs (65)-(414). ) And a triangle, so that G2 is <{((37), (65), (41)
4)), {(37)-(65), (37)-(414), (65)-(41
4)}> is initialized

[2: Addition of G2] Arcs (65)-(414) included in G2 are nodes (184) not included in G2.
And arcs (65)-(184) and arcs (184)-(414) form a triangle and are connected by a node (37) included in G2 and arcs (37)-(184). From the above, G2
, Nodes (184), arcs (37)-(184), arcs (65)-(184), and arcs (184)-(414) are added.

[3] Return to [2].

[2: Addition of G2] Since any arc included in G2 does not form a triangle with a node not included in G2, the process of G2 ends.

At this point, the following is performed. -------------------------------------------------- ----------- G0 = <{}, {}> G1 = <{(4), (37), (414), (502)}, {(4)-(37), (4)-(414), (37)-(414), (4)-(502), (37)-(502), (414)-(502)}> G2 = <{((37), (65), (184), (414)), {(37)-(65), (37)-(184), (37)-(414), (65)-(184), (65) -(414), (184)-(414)}> Arcs not belonging to any class: (3)-(37), (66)-(414), (66)-(974), (414) -(974) ---------------------------------------------- ---------------

[4] It is assumed that i = 2.

[1: Initial setting of G3] G0, G1, G
Arcs (68)-(414) not included in node 3
4) Arc (68)-(974), Arc (414)-(974)
And a triangle, G <2 (<{((68), (41)
4), (974)), {(68)-(414), (68)-(974), (41
4)-(974)}> is initialized.

[2: Addition of G3] Since any arc included in G3 does not form a triangle with a node not included in G3, the processing of G3 ends.

At this point, the following is performed. -------------------------------------------------- ----------- G0 = <{}, {}> G1 = <{(4), (37), (414), (502)}, {(4)-(37), (4)-(414), (37)-(414), (4)-(502), (37)-(502), (414)-(502)}> G2 = <{((37), (65), (184), (414)}, {(37)-(65), (37)-(184), (37)-(414), (65)-(184), (65) -(414), (184)-(414)}> G3 = <{((68), (414), (974)}, {(68)-(414), (68)-(974), ( 414)-(974)}> Arc not belonging to any class: (3)-(37) -------------------------- -----------------------------------

[4] It is assumed that i = 3.

[1: Initial setting of G4] G0, G1, G
2. Arcs (3)-(37) not included in G3 do not form a triangle with any node. Since there is no graph to be put in G4, the entire process is terminated.

As described above, when G0 is excluded, 3 shown in FIG.
One cluster is created. Here, "stadium"
(414) is common to G1, G2 and G3, and is a "goal"
Is common to G1 and G2.

[0137] Creation / Storage of Related Word Information After the processing in step 2 is completed, the results of word clustering are arranged and stored as related word information (step S3).

In order to obtain a related word of a specified word for each cluster, an affiliation cluster is obtained from the word, and
It is only necessary that words belonging to the cluster can be obtained.

The words belonging to the cluster are determined in step S2.
Can be obtained from the Gi node set obtained in (1). Table 6
Is a word table belonging to the cluster in the 10 words.

[0140]

[Table 6]

The table for obtaining the belonging cluster from the word is shown in Table 6.
Is obtained as shown in Table 7 by scanning using the word ID as a key.

[0142]

[Table 7]

In step 3, these tables are created and stored in the related word information storage unit 4 as a cluster dictionary and a word cluster dictionary, respectively.

Incidentally, in the word cluster dictionary, since words are actually specified by notations, the words are stored as shown in Table 8 using word notations as keys.

[0145]

[Table 8]

In displaying related words, Table 5 is also stored in the related word information storage unit 4 as a related degree dictionary in order to refer to the value of the related degree.

(B) Display of Related Word Information According to “(A) Creation / Storage of Related Word Information” described above, a user's instruction is issued while related word information of related word presentation device 1 is created / stored. "Display of related words" is performed.

Next, referring to the flowchart of FIG.
The flow of the process of “displaying related word information” of the related word presentation device 1 will be described.

[0149] Acceptance of Word Input The user accepts an input of a word for which a related word is to be obtained by the word input unit 5 (step S11). This implements a word input process.

. Acquisition of related word information Next, related word information of the word received by the word input unit 5 is acquired (step S12). This realizes a related word information acquisition step.

In the following, the word input section 5 uses the word “stadium”
Will be described with an example in which is input. That is, the input word “stadium” is searched in the word cluster dictionary storing Table 8. According to Table 8, “stadium” has a word ID of “41”.
4 "and three clusters" 1, 2, 3 ".

Next, according to the cluster dictionary storing Table 9,
The word ID of the word belonging to each cluster can be searched as follows.

Cluster 1: (4 words) 4,37,414,502 Cluster 2: (4 words) 37,65,184,414 Cluster 3: (3 words) 68,414,974

By obtaining the word (notation) of each word ID again from the word cluster dictionary of Table 8 except for the input word,
Related words can be obtained as follows.

Related words of cluster 1: 4 “J-League”,
37 “Goal”, 502 “Kashima” ・ Related words of Cluster 2: 37 “Goal”, 65 “Try”,
184 “Hanazono” ・ Related words of Cluster 3: 68 “Truck”, 974 “Athletics”

Further, the degree of association with the input word may be obtained. The relevance is obtained from the relevance dictionary in Table 5. For example, the degree of association between “stadium” having a word ID of 414 and “J-League” having a word ID of 4 is the value “5.69” of the element of (4,414) on the matrix of the degree of association dictionary. As a result, the following degree of association (in []) can be added.

Related words for cluster 1: 4 "J-League"
[5.69], 37 "goals" [4.93], 502 "Kashima" [5.29]-Related words of cluster 2: 37 "goals" [4.93], 65
"Try" [4.04], 184 "flower garden" [4.38] ・ Related words of cluster 3: 68 "truck" [5.07], 974
"Athletics" [5.18] Although not shown in the above example, the cluster obtained here is not a complete graph, so some related words may not have a direct relevance to the word.

[0158] Those below a certain threshold may be excluded from related words. For example, if the threshold is set to 4.5, the following is obtained.

Related words for cluster 1: 4 "J-League"
[5.69], 37 "goals" [4.93], 502 "Kashima" [5.29]-Related words of cluster 2: 37 "goals" [4.93]-Related words of cluster 3: 68 "tracks" [5.07], 974
"Athletics" [5.18]

Further, an index based on the degree of association between a word and a word in a cluster may be used. For example, assuming that the average value of the relevance to the input word is used as the index, the index is as follows.

Cluster 1: 5.30 Cluster 2: 4,45 Cluster 3: 5.13

. Display of related word information Then, related words of the input word obtained in step S2 are grouped for each cluster and displayed (step S13). This realizes a related word information output step.

For example, the following is displayed. -------------------------------------------------- ----------- [A]: League goal Kashima [B]: Goal Try Hanazono [C]: Track athletics ------------------ ------------------------------------------- In step S2, below a certain threshold When excluding things, they are also excluded from the display. For example, if the threshold value is 5, the display is as follows. -------------------------------------------------- ----------- [A]: League Kashima [B]: Track Athletics -------------------------- ----------------------------------- In addition, it is displayed in the order of the degree of relevance as follows You may make it. -------------------------------------------------- ----------- [A]: Goal Kashima goal [B]: Goal Hanazono try [C]: Athletics track ------------------ ------------------------------------------- In step S2, words and clusters When the index based on the relevance of the word is used, clusters are displayed in the order of the index based on the relevance of the word and the word in the cluster. For example, assuming that the average value of the relevance to the input word is used as the index, the following is displayed. -------------------------------------------------- ----------- [A]: League goal Kashima [B]: Track athletics [C]: Goal Try Hanazono ------------------ -------------------------------------------

[Second Embodiment of the Invention] Another example of the embodiment of the present invention will be described as a second embodiment of the present invention.

FIG. 7 is a functional block diagram of the document search device 21 according to the second embodiment of the present invention. As shown in FIG. 7, the search sentence input unit 22 receives an input of a search sentence for document search, sends the search sentence to the search unit 23, and also receives an input of a related word presentation instruction. The search sentence is also sent to the related word presentation unit 24.

The document database unit 25 stores a document database for registering and managing a plurality of documents, and is managed by, for example, a relational database management system. The information managed by this database includes at least the title and document content information of one document.

The search unit 23 searches the document database using the search sentence provided from the search sentence input unit 22, and sends the search result to the search result output unit 26. In response, the search result output unit 26 outputs the search result.

The related word presentation section 24 extracts words in the search sentence provided from the search sentence input section 22, acquires related words, and outputs the obtained result to the related word information output section 7 (see FIG. 8). At the same time, the information of the related word is also sent to the search sentence correcting unit 27.

The search sentence correcting unit 27 adds all of the related words presented by the related word presentation unit 24 or the related words specified by the input by the search sentence input unit 22 to, for example, a search sentence. Modify the search sentence.

FIG. 8 is a functional block diagram of the related word presentation unit 24. As shown in FIG. 8, the related word presentation unit 24 includes a word clustering unit 3, a related word information storage unit 4, a word input unit 5, similar to the related word presentation device 1 according to the first embodiment of the present invention.
A related word information acquisition unit 6 and a related word information output unit 7 are provided. The document database unit 25 is used as the document database unit 2. Further, a word acquisition unit 28 is provided instead of the word input unit 5. The word acquisition unit 28 extracts a word in the search sentence input by the search sentence input unit 22 to acquire a related word, and outputs the obtained result to the related word information output unit 7 and the related word information acquisition unit 6. In this case, if the search sentence correction unit 27 corrects the search sentence only for the related words specified by the input by the search sentence input unit 22, only the related words specified by the input by the search sentence input unit 22 Is output to the related word information acquisition unit 6.

The specific hardware configuration of the document search device 21 shown in FIGS. 7 and 8 is the same as that of the related word presentation device 1 according to the first embodiment of the present invention shown in FIG. A document search program is installed on the hard disk 17. This document search program is stored in the storage medium 18 and can be read by the storage medium reading device 19 and installed on the hard disk 17. Also, a related word presentation program is downloaded from a network such as the Internet,
7 can be installed.

The document database section 25 is constructed in the hard disk 17. When the document search program is installed on the hard disk 17, the CPU 11
Executes the functions of the search unit 23, the related word presentation unit 24, and the search sentence correction unit 27 based on the document search program. The storage area of the related word information storage unit 4 is secured in the RAM 13 or the hard disk 17. The function of the word input unit 5 is executed via the input device 16, and the functions of the search result output unit 26 and the related word information output unit 7 are executed via the output device 15. The document search program may operate on a predetermined OS. Further, a part of the function may be performed on behalf of the OS or other application software.

Next, the processing performed by the document search device 21 will be described with reference to the flowcharts of FIGS. As shown in FIG. 9, first, a search sentence input for a required document is received from the user via the search sentence input unit 22 (step S21). This implements a search sentence input step. Next, a search term, a search condition expression in which the search term is combined with a logical symbol such as an AND operator (logical product) or an OR operator (logical sum), or a part of a document composed of natural language sentences or natural language sentences The search unit 23 performs a search on the document database unit 25 under the condition that the search sentence is all or all (step S22), and displays the search results as a list of document titles via the search result output unit 26 (step S23). . The search step is realized by step S22, and the search result output step is realized by step S23.

In the flow of the document search in steps S21 to S23, at various occasions such as during input of the search text, after input, after browsing of the search result, the request for the related word presentation is input. When received from the user via the unit 22 (step S31), a word in the search sentence is extracted (step S32). Then, related word information is obtained for each word in the search sentence (step S33), and the obtained related word information is output by the related word information output unit 7. Then, a correction instruction is received from the user for the output related word information (step S34), and the search text is corrected (step S3).
5). A related word presentation process is realized by steps S31 to S33, and a word acquisition process is realized by step S32.
Steps S34 and S35 implement a search sentence correcting step.

[0175]

According to the first aspect of the present invention, it is possible to present related words for each cluster having a high degree of mutual relevance by using the created related word information, so that the user can efficiently obtain related words. We can help you make a choice.

According to a second aspect of the present invention, in the related word information creating apparatus according to the first aspect, the related words are mutually related by using the related word information created based on the degree of relevance between words. It is possible to present for each high cluster, and it is possible to assist the user in efficiently selecting related words.

According to a third aspect of the present invention, in the related word information creating apparatus according to the second aspect, non-exclusive clustering is performed, and related words can be presented for each cluster having a high degree of mutual relation. As a result, it is possible to assist the user in efficiently selecting related words.

According to a fourth aspect of the present invention, in the related word information creating apparatus according to any one of the first to third aspects, the related words are related to each other by referring to the cluster dictionary and the word cluster dictionary. It is possible to present for each cluster having a high level, and to assist the user in efficiently selecting related words.

The invention according to claim 5 makes it possible to present related words for each cluster having a high degree of mutual relevance by using the created related word information, thereby enabling the user to efficiently select related words. I can help.

According to a sixth aspect of the present invention, in the related word presenting apparatus according to the fifth aspect, only a word having a high degree of relevance to a word for which a related word is to be obtained can be obtained as a related word.

According to a seventh aspect of the present invention, in the related word presenting apparatus according to the fifth aspect, only a word having a high degree of relevance to a word for which a related word is to be obtained can be output as a related word.

[0182] According to the invention of claim 8, in the related word presentation device according to any one of claims 5 to 7, the related words are displayed in the order of indices based on the degree of relevance to the word whose related word is to be obtained. Can be output.

According to the ninth aspect of the present invention, it is possible to present related words of words included in a search sentence for each cluster having a high degree of mutual relevance, thereby preventing search omission and improving the efficiency of search narrowing. Can be improved.

According to the tenth aspect of the present invention, it is possible to present related words for each cluster having a high degree of mutual relevance by using the created related word information, thereby enabling a user to efficiently select related words. I can help.

According to an eleventh aspect of the present invention, in the related word information creating method according to the tenth aspect, the related words are mutually related by using the related word information created based on the degree of association between the words. It is possible to present for each high cluster, and it is possible to assist the user in efficiently selecting related words.

According to the twelfth aspect of the present invention, by performing non-exclusive clustering, related words can be presented for each cluster having a high degree of mutual relevance, and the user can efficiently select related words. I can help.

The thirteenth aspect of the present invention relates to the tenth aspect.
12. In the related-word information creating method according to any one of the above 12, the cluster word dictionary and the word cluster dictionary are referred to,
Related words can be presented for each cluster having a high degree of mutual relevance, and it is possible to assist the user in efficiently selecting related words.

The invention according to claim 14 makes it possible to present related words for each cluster having a high degree of mutual relevance by using the created related word information, so that the user can efficiently select related words. I can help.

According to a fifteenth aspect of the present invention, in the related word presentation method according to the fourteenth aspect, only a word having a high degree of relevance to a word for which a related word is to be obtained can be obtained as a related word.

According to a sixteenth aspect of the present invention, in the related word presentation method according to the fourteenth aspect, only a word having a high degree of relevance to a word whose related word is to be obtained can be output as a related word.

The invention according to claim 17 is the invention according to claims 14 to
16. The related-word presentation method according to any one of 16.
Related words can be output in the order of indices based on the degree of relevance to the word for which a related word is to be obtained.

The invention according to claim 18 makes it possible to present related words of a word included in a search sentence for each cluster having a high degree of mutual relation, thereby preventing search omission and improving the efficiency of search narrowing. Can be improved.

According to the nineteenth aspect of the present invention, it is possible to present related words for each cluster having a high degree of mutual relevance by using the created related word information, so that the user can efficiently select related words. I can help.

According to a twentieth aspect of the present invention, there is provided the storage medium according to the nineteenth aspect, wherein related words are created for each cluster having a high degree of mutual relevance by using related word information created based on the degree of relevancy between words. To support the user in efficiently selecting related words.

According to the twenty-first aspect, in the storage medium according to the twentieth aspect, non-exclusive clustering is performed so that related words can be presented for each cluster having a high degree of mutual relation. Can help people to select related words efficiently.

The invention according to claim 22 is the invention according to claims 19 to
21. The storage medium according to any one of 21., wherein the related word can be presented for each cluster having a high degree of mutual relevance by referring to the cluster dictionary and the word cluster dictionary. Can help you make better choices.

The invention described in claim 23 is the invention according to claims 19 to
22. In the storage medium according to any one of 22., it is possible to present related words for each cluster having a high degree of mutual relevance by using the created related word information, so that a user can efficiently select related words. Can help.

According to the twenty-fourth aspect of the present invention, in the storage medium according to the twenty-third aspect, only those having a high degree of relevance to a word for which a related word is to be obtained can be obtained as related words.

According to the twenty-fifth aspect of the present invention, in the storage medium according to the twenty-third aspect, only a word having a high degree of relevance to a word whose related word is to be obtained can be output as a related word.

The invention according to claim 26 is the invention according to claims 23 to
25. The storage medium according to any one of 25, wherein the related words can be output in the order of the index based on the degree of relevance to the word for which the related word is to be obtained.

The invention according to claim 27 is the invention according to claims 23 to
26. In the storage medium according to any one of the items 26, it is possible to present related words of a word included in the search sentence for each cluster having a high degree of mutual relevance, thereby preventing search omission,
The efficiency of search narrowing can be improved.

[Brief description of the drawings]

FIG. 1 is a functional block diagram of a related word presentation device according to a first embodiment of the present invention.

FIG. 2 is a block diagram showing a hardware configuration of the related word presentation device.

FIG. 3 is a flowchart showing a process performed by the related word presentation device.

FIG. 4 is an explanatory diagram illustrating creation of related word information by the related word presentation device.

FIG. 5 is an explanatory diagram illustrating creation of the related word information.

FIG. 6 is a flowchart showing a process performed by the related word presentation device.

FIG. 7 is a functional block diagram of a document search device according to a second embodiment of the present invention.

FIG. 8 is a functional block diagram of a related word presentation unit of the document search device.

FIG. 9 is a flowchart illustrating a process performed by the document search device.

FIG. 10 is a flowchart of the same.

[Explanation of symbols]

 REFERENCE SIGNS LIST 1 related word presentation device 2 document database unit 3 word clustering unit 4 related word information storage unit 5 word input unit 6 related word information acquisition unit 7 related word information output unit 18 storage medium 21 document search device 22 search text input unit 23 search unit 24 related word presentation unit 25 document database unit 26 search result output unit 27 search sentence correction unit 28 word search unit

Claims (27)

[Claims] 1. A related word information creating apparatus for creating related word information to be referred to to acquire a related word of a given word, wherein the related word information creating device refers to a document database unit that stores a plurality of documents.
A word clustering unit that clusters words included in the plurality of documents, classifies the words into clusters for each meaning or field, and creates the related word information. . 2. The related word information creating apparatus according to claim 1, wherein the word clustering unit performs the clustering based on a degree of association between the words. 3. The word clustering unit extracts a highly relevant subgraph as a cluster from the entire graph in which the word is a node and the relation between words is an arc, and obtains a node set of the obtained cluster. 3. The related-word information creating apparatus according to claim 2, wherein the clustering is performed. 4. The word clustering unit includes: a cluster dictionary that associates a cluster ID for identifying the cluster with a word ID for identifying the word belonging to the cluster; and for each word, the word ID of the word and The related word information creating apparatus according to any one of claims 1 to 3, wherein a word cluster dictionary in which a cluster ID of the cluster to which a word belongs is created as the related word information. 5. A related-word information creating device according to claim 1, a related-word information storage unit that stores the related-word information created by the related-word information creating device, and a related word. A word input unit for receiving an input of a word to obtain a related word, a related word information acquiring unit for referring to the related word information storage unit to find a related word of a given word, and a related unit for outputting the obtained related word. And a word information output unit. 6. The related word information acquiring unit restricts only words that can be obtained as the related words by referring to the related word information storage unit to those having a high degree of relevance to the word input in the word input unit. The related word presentation device according to claim 5, wherein the related word is obtained as the related word. 7. The related word information output unit outputs only the related words obtained by the related word information acquisition unit that have a high degree of relevance to the word input by the word input unit. The related word presentation device according to claim 5, wherein 8. The related word information output unit outputs the related words in the same cluster in the order of indices based on the degree of relevance to the word input by the word input unit. Item 8. The related-word presentation device according to any one of Items 5 to 7. 9. A search sentence input unit for receiving an input of a search sentence, a search unit for searching for a document to be matched with the input search sentence by referring to a document database unit storing a plurality of documents, A search result output unit for outputting the obtained search result, a word is extracted from the search sentence input by the search sentence input unit instead of the word input unit, and the extracted word is used as the related word information acquisition unit The related word presentation device according to any one of claims 5 to 8, further comprising: a word acquisition unit configured to input to the related word obtained by the related word presentation device based on all or specified words. And a search sentence correction unit for correcting the search sentence input by the search sentence input unit based on the search sentence input unit. 10. A related word information creating method for creating related word information to be referred to to obtain a related word of a given word, wherein the plurality of documents are registered by referring to a document database unit. A related word information creating method, comprising: clustering words included in a document, classifying the words into clusters for each meaning or field, and creating the related word information. 11. The related word information creating method according to claim 10, wherein in the word clustering step, the clustering is performed based on a degree of association between the words. 12. The word clustering step extracts a complete graph or a transitory graph slightly loosened as a cluster from the entire graph in which the word is a node and the relationship between words is an arc, and the obtained cluster is obtained. 12. The related-word information creating method according to claim 11, wherein the clustering is performed by obtaining a set of nodes. 13. The word clustering step includes a cluster dictionary in which a cluster ID for identifying the cluster and a word ID for identifying the word belonging to the cluster are associated with each other. 13. The related word information creating method according to claim 10, wherein a word cluster dictionary that associates the cluster ID of the cluster to which a word belongs with is created as the related word information. 14. A word input step for receiving an input of a word for which a related word is to be acquired, and the related word information created by the related word information creating method according to claim 10. A related word information obtaining step of referring to a related word information storage unit to obtain a related word of the word given in the word input step; a related word information output step of outputting the obtained related word;
A related word presentation method characterized by comprising: 15. The related word information acquiring step is limited to a word which can be obtained as the related word by referring to the related word information storage unit, to a word having a high degree of relevance with the word input in the word input unit. The related word presentation method according to claim 14, wherein the related word is obtained as the related word. 16. The related word information output step includes, of the related words obtained by the related word information acquiring unit, outputting only the related words having a high degree of relevance to the word input by the word input unit. The related word presentation method according to claim 14, wherein the related word presentation method is used. 17. The related word information output step, wherein the related words in the same cluster are output in the order of indices based on the degree of relevance to the word input in the word input unit. Item 17. The related word presentation method according to any one of Items 14 to 16. 18. A search sentence input step of receiving a search sentence input, a search step of searching for a document to be matched with the input search sentence by referring to a document database unit storing a plurality of documents, A search result output step of outputting the obtained search result; and, in place of the word input step, extracting a word from the search sentence input in the search sentence input unit, and extracting the extracted word into the related word information obtaining step. A related word presenting step of obtaining the related word based on all or specified words, by the related word presenting method according to any one of claims 14 to 17, comprising: A search sentence correcting step of correcting the search sentence input in the search sentence input step based on the obtained related words. 19. A computer-readable storage medium storing a related word information creating program for causing a computer to create related word information to be referred to in order to acquire a related word of a given word, The word information creation program clusters words included in the plurality of documents by referring to a document database unit in which the plurality of documents are registered, classifies the words into clusters for each meaning or field, and generates the related word. A storage medium for causing a computer to execute a word clustering step of creating information. 20. The storage medium according to claim 19, wherein in the word clustering step, the clustering is performed based on a degree of association between the words. 21. The word clustering step extracts, as a cluster, a complete graph or a transitivity graph slightly loosened from the entire graph in which the word is a node and the relationship between the words is an arc. 21. The storage medium according to claim 20, wherein the clustering is performed by obtaining a set of nodes. 22. The word clustering step includes: a cluster dictionary in which a cluster ID for identifying the cluster is associated with a word ID for identifying the word belonging to the cluster; 22. The storage medium according to claim 19, wherein a word cluster dictionary in which a cluster ID of the cluster to which a word belongs is created as the related word information. 23. A word input step of receiving an input of a word for which a related word is to be acquired, and referring to a related word information storage unit storing the related word information created by a computer based on the related word information creation program. A related word information obtaining step of obtaining a related word of the word given in the word input step; and a related word information output step of outputting the obtained related word. The storage medium according to any one of claims 19 to 22, wherein the storage medium is a storage medium. 24. The related word information acquiring step is limited to a word that can be obtained as the related word by referring to the related word information storage unit, to a word having a high degree of relevance with the word input in the word input unit. The storage medium according to claim 23, wherein the storage medium is determined as the related word. 25. The related word information output step includes, of the related words obtained by the related word information obtaining unit, outputting only the related words having a high degree of relevance to the word input by the word input unit. The storage medium according to claim 23, characterized in that: 26. The related word information output step, wherein the related words in the same cluster are output in the order of indices based on the degree of relevance to the word input by the word input unit. Item 30. The storage medium according to any one of items 23 to 25. 27. A search sentence input step of receiving a search sentence input, a search step of searching for a document to be matched with the input search sentence by referring to a document database unit storing a plurality of documents, A search result output step of outputting the obtained search result; and, in place of the word input step, extracting a word from the search sentence input in the search sentence input unit, and extracting the extracted word into the related word information obtaining step. By the word acquisition step to be provided and the related word presentation program to be executed by the computer,
A related word presenting step of obtaining the related word based on all or specified words; a search sentence correcting step of correcting the search sentence input in the search sentence input step based on the obtained related word; The storage medium according to any one of claims 23 to 26, which stores a document search program for causing a computer to execute the following.