JP2007004831A - Document retrieval system using word meaning-related network - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To dissolve polysemy of a retrieval word entered by a user, and to retrieve a related document by using a general search engine. <P>SOLUTION: This document retrieval system has: a word meaning-related network exhibition part 301 exhibiting a word meaning of a retrieval key entered by the user and word meanings related to the word meaning like a network; a retrieval part 302 producing a retrieval key from the word meaning selected by the user and executing the retrieval; and a filtering part 304 selecting only the documents suitable for the word meanings from a result obtained by executing the retrieval. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a document retrieval system that retrieves a related document using a meaning representing a user's retrieval intention.

  With the spread of computers and the Internet, computerization of document information is rapidly progressing. As available document information increases, it becomes an important issue to search for necessary information. Information retrieval technology (also referred to as document retrieval technology) is a technology for solving such problems, and it is a common practice to retrieve necessary document information using a search engine that applies that technology. ing. Further, in recent years, there is an increasing demand for a cross-lingual document search technique (also referred to as a cross-language document search technique) for searching a document described in a language different from a search key input by a user.

  In a monolingual document search in which a document described in the same language as the search key input by the user is searched, the search target document is searched using the search key input by the user. On the other hand, in the cross-lingual document search, since the search key input by the user and the search target document are described in different languages, it is necessary to translate the search key into the language of the search target document. For example, in the case of English-Japanese cross-lingual document search that searches a document written in Japanese with respect to a search key written in English, the search key (for example, “camera”) entered by the user in English is used in Japan. A Japanese document is searched after translation into a word search key (for example, “camera”).

JP 2001-331484 A

  However, since a word generally has a plurality of meanings (meanings), it is not always possible to uniquely translate a search key into a search target language. For example, consider the case of translating English “bank” into Japanese. “Bank” has a meaning as an institution that deposits and withdraws money (corresponding to “bank” in Japanese) and waterside. Since there is a meaning (corresponding to “Kishi” in Japanese) as the place of the place, it is not clear which one should be translated, and it may be translated into a meaning that the user does not intend. As a result, documents unintended by the user are included in the search results.

  A similar problem occurs in monolingual document retrieval. When the search key input by the user includes a plurality of meanings, a document related to the meaning not intended by the user is obtained as a search result. For example, if you enter English “tank” as a search key, “tank” has a meaning equivalent to Japanese “tank” or “aquarium” and a meaning equivalent to “tank”. , When performing a monolingual document search using “tank” as a search key, even if the user intends to search for the meaning of the Japanese “tank”, Documents related to meaning are included in the search results.

  In this way, when the search key has a plurality of meanings, if the user does not identify the meaning of the plurality of meanings as the search intention, a document search reflecting the user's search intention is performed. I can't.

  An object of the present invention is to provide a document search system that can solve the ambiguity of a search key and obtain a search result reflecting a user's search intention.

  In order to achieve the above object, the present invention presents a meaning-related network in which the meaning of a search key input by a user and the meaning related to the meaning are arranged in a network, and the meaning that reflects the search intention from the network. The user's search intention is clarified by letting the user select.

  Here, the meaning displayed in the meaning-related network is defined as a set of words (terms) in different languages (for example, English and Japanese). For example, the meaning of English “bank” is “bank” as an institution for depositing and withdrawing money and “bank” as a waterside place. In this case, “bank” and “bank” are defined as the meaning of “bank”. A network structure in which the meaning defined in this way is a node and the degree of association between meanings is a link is a meaning-related network. The method for creating the semantic network may be arbitrary. For example, Japanese Patent Application Laid-Open No. 2001-331484 "Recording medium recording a parallel thesaurus generation program, recording medium recording a parallel thesaurus, and recording medium recording a parallel thesaurus navigation program" Can be used. The parallel thesaurus referred to in this method corresponds to the meaning-related network in the present invention.

  The user's search intention is clarified by selecting the meaning from the meaning-related network. However, typical search engines (Google (http://www.google.com), goo (http://www.goo.ne.jp), etc.) that have a keyword search function are used in the present invention. And they accept words (terms), not meanings. That is, it is impossible to search using the meaning specified by the user as a key. Therefore, the search must be executed using the term before identifying the meaning as a key, and a document not related to the meaning specified by the user is included in the search result. Therefore, in the present invention, the meaning of terms included in the search result document is determined, and filtering is performed by selecting only documents in which the terms are used in the meaning specified by the user. Search results that reflect the search intentions of the user.

  That is, a document search system according to the present invention for searching for a document related to an input search key includes a meaning-related network presentation unit that presents the meaning of the search key and the meaning related to the meaning in a network form, and the search key. And a search unit that executes a search using a search unit, and a filtering unit that selects a document that matches the selected meaning from a set of documents obtained by executing the search. The meaning-related network presentation unit presents the meaning in a network form in a format that prompts the user to select a meaning.

  The search unit may execute a search using the input search key, or may generate a search key from the selected meaning and execute the search using the generated search key. In addition, the search unit may perform a cross-lingual document search for searching for a document described in a language different from the input search key, or a document described in the same language as the input search key. A monolingual document search for searching may be performed.

  According to the present invention, a search result reflecting a user's search intention can be obtained by selecting a meaning displayed on the meaning-related network and executing a search.

A typical aspect of the present invention includes (1) a monolingual document search for searching a document described in the same language as the search key input by the user, and (2) a language different from the search key input by the user. There is a cross-lingual document search, which searches the described document. The language of the search key and the language of the document to be searched may be a combination of arbitrary languages. Hereinafter, as an example of the above (1), an English document is searched for an English search key, and as an example of the above (2), an English document is searched for a Japanese search key. The case of searching for will be described.
Hereinafter, the embodiment (1) will be described with reference to the drawings.

  FIG. 1 is a schematic diagram showing a configuration example of a system for realizing the present invention. This system includes a client 20, a server 30, a meaning-related database 40, and search engines 501 and 502, and the client 20, server 30, and search engines 501 and 502 are connected via a communication network 10. In the example shown in the figure, two search engines are connected to the communication network, but the number of search engines used may be arbitrary. The number of clients is also arbitrary.

  The client 20 includes a search request input unit 201, a search engine selection unit 202, a meaning-related network display unit 203, and a search result display unit 204. The function of each part will be described later.

  The server 30 includes a meaning-related network generation unit 301, a search unit 302, a meaning determination unit 303, and a filtering unit 304. The function of each part will be described later.

  The meaning-related database 40 is a database that stores information about meaning, (a) a table that holds information about the meaning of a certain word (term), (b) a table that holds information about the degree of association between meaning and meaning, (C) When a certain term 1 co-occurs with another term 2, a table is provided for holding information about which term 1 is likely to have a meaning.

  FIG. 2 is a diagram illustrating an example of a table provided in the meaning-related database 40. (A), (b), and (c) in FIG. 2 correspond to the above (a), (b), and (c), respectively. FIG. 2A shows that the term “tank” has two meanings “tank / tank / tank” and “tank / tank”. Fig. 2 (b) shows that the relationship between the meaning "tank tank / aquarium" and the meaning "ammonia ammonia" is 4.834, and the meaning between the meaning "tank tank" and the meaning "artillery gun" is 4.044. It means that it is. Figure 2 (c) shows that when the term “tank” co-occurs with the term “ammonia”, the term “tank / tank” is more likely to be used than the term “tank / tank”. (It can be found by comparing 0.169 and 4.834 with the meaning of meaning), and when it co-occurs with the term “artillery”, it is easier to have the meaning of “tank / tank” than the meaning of “tank / tank” (It can be found by comparing 0.063 and 4.044 for each word meaning). This table is used when the meaning determination unit 303 determines the meaning that the term of interest takes in the document. The function of the meaning determination unit 303 will be described later.

  The search engines 501 and 502 are search engines having a keyword search function, and general search such as Google (http://www.google.com) and goo (http://www.goo.ne.jp). The engine can be used.

  Hereinafter, with respect to the function of each unit, FIG. 3 is a sequence diagram showing the flow of data and processing between the units, FIG. 4, FIG. 5, FIG. This will be described in detail with reference to FIG. 2 showing an example of the included table.

  FIG. 4 is a diagram illustrating an example of an initial screen in the client 20. The search request input unit 201 includes a search request input area 2011 and a meaning-related network display instruction button 2012. For example, when “tank” is input as an English search key in the search request input area 2011 and the meaning-related network display instruction button 2012 is clicked, the input search key “tank” is the meaning-related network generation unit 301 of the server 30. (T1 in FIG. 3).

  The meaning-related network generation unit 301 searches the meaning of the search key “tank” and the meaning (related meaning) having a high degree of association with the meaning using the meaning-related database 40 (a) and (b) (see FIG. 3 T2, T3), a meaning-related network is generated. The number of related meanings to be searched may be arbitrary, but about 5 is reasonable.

  The generated meaning-related network is transmitted to the client 20 (T4 in FIG. 3) and displayed on the meaning-related network display unit 203 of the client 20 as shown in FIG. By looking at the displayed meaning-related network, the user can see that the search key “tank” has two meanings “tank tank / tank” and “tank tank”, and the meaning “tank tank”. / Aquarium is related to terms such as "ammonia / ammonia" and "fuel / fuel", and the meaning of "tank / tank" is related to terms such as "artillery" and "ball" You can know what to do. The user can use the meaning selection button 2031 in the meaning-related network display unit 203 to select the meaning that the user wants to search from the meanings of the search key. Further, by using the related meaning selection button 2032, it is possible to select the related meaning to be searched together with the meaning of the search key.

  Next, the user uses the search engine selection unit 202 provided in the meaning-related network display unit 203 to select a search engine to be searched with the search engine selection button 2022 and clicks the search instruction button 2021. The information on the selected meaning and related meaning is transmitted to the search unit 302 of the server 30 (T5 in FIG. 3). At the same time, the name of the search engine selected as the search target is also transmitted.

  The search unit 302 generates a search key that can be accepted by the search engine based on the transmitted meaning. For example, the search key “tank” is generated from the meaning “tank tank / water tank”. A search engine is searched using the generated search key (T6 in FIG. 3), and a document related to the search key is obtained as a result (T7 in FIG. 3). The obtained document set is not a search by the meaning “tank / tank / aquarium” intended by the user, but is a result of a search by the term “tank”, and therefore includes a result that does not satisfy the user's intention. ing. That is, it is necessary to select a document in which the term “tank” is used in the meaning intended by the user from the document set obtained as a search result. As preprocessing for document selection, the meaning determination unit 303 determines in which meaning the term “tank” included in the document obtained as a search result is used.

  The meaning determination unit 303 performs the meaning determination using the table (c) of the meaning-related database 40 (T8 and T9 in FIG. 3). For example, if the term “tank” co-occurs with the term “ammonia” in a document obtained as a search result, the term “tank” in this document is used in the meaning of “tank / tank / aquarium”. (It can be found by comparing the relevance values of 0.169 and 4.834). Similarly, for all terms co-occurring with the term “tank”, the meaning of the term “tank” is determined in consideration of the degree of association with each term. This meaning determination is executed for each document obtained as a search result.

  The filtering unit 304 selects only documents in which “tank” is used in the meaning satisfying the user's search intention from among the document sets whose meaning has been determined, and the document set obtained as a result is sent to the client 20. Transmit (T10 in FIG. 3).

  The transmitted document set is displayed on the search result display unit 204 in the client 20. FIG. 6 is a diagram illustrating an example of search result display. The display format of the search result may be simply the title list of the document obtained as the search result. However, as shown in FIG. 6, in addition to the document title, a part of the text is displayed and displayed in the displayed text. The search key and the term used for the meaning determination may be highlighted. By doing so, in FIG. 6, the search key “tank” is italicized, and the terms “ammonia”, “fuel”, and “toilet” used for meaning determination are highlighted in bold.

  By using the system described above, even when using a general search engine, a document in which the search key is used in the meaning intended by the user from among the related documents for the search key entered by the user Can be obtained.

  Note that the search unit 302 of the server 30 does not use the search key generated based on the meaning information transmitted from the client 20 in the search of the search engine (T6 in FIG. 3). The search engine may be searched by using the search key (T1 in FIG. 3) transmitted to. Even in this case, the meaning determination unit 303 determines the meaning of the search key included in each document for the document set obtained by the search, and the filtering unit 304 uses the search key with the meaning intended by the user. A search result reflecting the user's search intention can be obtained by performing processing for selecting a document that has been read.

  In the above, the embodiment in the case of searching for an English document with respect to an English search key (monolingual document search) has been described. Next, as a modification of the present invention, a case where an English document is searched with respect to a Japanese search key (cross-lingual document search), which is an embodiment of the aspect (2), will be described.

  In order to realize the embodiment of the aspect of (2), a table storing information related to the meaning of Japanese terms is required as the table provided in the meaning-related database 40. For example, information such as “tank / tank” for the Japanese term “tank” and “fuel / fuel” for the Japanese term “fuel” may be stored. Such a table can be created in the same manner as the table (a) included in the meaning-related database 40 in the embodiment of the aspect (1).

  The flow of processing in the embodiment of the aspect of (2) is almost the same as that of the embodiment of the aspect of (1), but the search key (for example, “tank”) entered by the user by the meaning-related network generation unit 301. When searching for the meaning of the word, the table storing the information about the meaning of the Japanese term above may be searched. Then, the meaning (for example, “tank / tank”) for the search key is obtained. Thereafter, the same processing as in the embodiment of the aspect (1) is executed.

  Advantages of the embodiment of the aspect (2) are as follows. In a general cross-lingual document search, an input search key is translated in a bilingual dictionary and then a document is searched. For example, when a Japanese search key “tank” is input, the search key is translated into “tank” using a bilingual dictionary, and a document is searched using “tank”. However, as mentioned in the above example, “tank” includes multiple meanings, so documents containing “tank” with meanings other than “tank” in the meaning of the tank intended by the user are searched. It will be done. However, in the embodiment of the aspect (2), since the meaning of “tank” is identified as “tank / tank”, the intention of the user is not affected by the ambiguity of “tank”. In the same way, a document in which “tank” is used to mean “tank” can be presented to the user.

1 is a schematic diagram showing a configuration example of a system for realizing the present invention. The figure which shows the example of the table contained in a meaning-related database. The sequence diagram which shows the data between a client, a server, a meaning-related database, and a search engine, and the flow of a process. The figure which shows the example of the initial screen in a client. The figure which shows the example of a display of the meaning related network in a client. The figure which shows the example of a display of the search result in a client.

Explanation of symbols

10: communication network 20: client 201: search request input unit 2011: search request input area 2012: meaning-related network display instruction button 202: search engine selection unit 2021: search instruction button 2022: search engine selection button 203: meaning-related network display Unit 2031: Meaning selection button 2032: Related meaning selection button 204: Search result display unit 30: Server 301: Meaning related network generation unit 302: Search unit 303: Meaning determination unit 304: Filtering unit 40: Meaning related database 501, 502: Search engine

Claims (6)

In a document search system that searches for documents related to a search key,
A database that stores terms and their meanings, the degree of association between meanings and meanings, and the degree of association between terms and terms;
A search key input unit for inputting a search key;
The meaning of the search key and the meaning related to the meaning are extracted from the database from the input search key, and the meaning-related network presentation unit presenting the meaning in a network form;
Meaning selection means for selecting any one or more of the meanings presented in the meaning-related network presenting unit;
A search unit that performs a search using a search key generated based on the meaning selected by the meaning selection means;
For each document of the document set obtained by performing a search by the search unit, using the database, a meaning determination unit that determines in which sense a term in the document is used,
A document search system comprising: a filtering unit that selects a document in which the meaning selected by the meaning selection means is used from the document set determined by the meaning determination unit.   The document search system according to claim 1, wherein the search unit generates a search key from the selected meaning and performs a search using the generated search key.   The document search system according to claim 1, wherein the search unit performs a cross-lingual document search for searching for a document described in a language different from the input search key.   2. The document search system according to claim 1, wherein the search unit performs a monolingual document search for searching for a document described in the same language as the input search key.   3. The document search system according to claim 2, wherein the search unit performs a cross-lingual document search for searching for a document described in a language different from the input search key.   3. The document search system according to claim 2, wherein the search unit performs a monolingual document search for searching for a document described in the same language as the input search key.

Images