JPH03130873A - Text base retrieving system - Google Patents

Abstract

PURPOSE:To improve the retrieving accuracy of a text base by choosing the words to be retrieved out of the input sentences to evolve these words to the similar characters to the component KANJI (Chinese character) and adding an obtained character string to a retrieving character string. CONSTITUTION:A text base 8 consisting of the documents of a natural language is retrieved, and a text to be retrieved is designated by the natural language or a word (key word). The word used in an input sentence is retrieved with a word evolved to the KANJI showing the meaning similar to each KANJI forming the relevant word. Then the expressions having the similar meanings can be retrieved out of a retrieving structure with use of a similar character dictionary 6 when the produced retrieving structure is reinforced. Thus the expression similar to a word used in an input sentence can be retrieved and therefore the retrieving accuracy of the base is improved.

Description

[Detailed Description of the Invention] [Industrial Application Field] The IJJ of this invention is based on natural language makeshift sentences or keywords in a database (hereinafter referred to as text base) that stores natural language text data as character code strings. Regarding the text-based search method that searches for sentence 0 data containing synonymous expressions, it is especially possible to search for words used in the input sentence and synonymous expressions, improving search accuracy. Regarding.

[Conventional technology]

Conventionally, as a text-based search method, for example, “
Sugiyama et al., IRIS, an information retrieval system based on natural language;
Information Processing Society of Japan Natural Language Processing Study Group Materials NL-58-8,
1986, by assigning keywords (fields or words) appropriate to each text as data, the content or characteristics of each text can be expressed, and when searching. The user specifies keywords related to the content of the text desired and their logical relationships (AND, OR, etc.) and extracts texts that satisfy the conditions.

In such search methods, the phrases that come to mind when a user makes a search request are not necessarily used exactly in the text. For example, it is necessary to deal with synonymous expressions, such as paraphrasing using synonyms such as ``a foothill person dressed as a man'' and ``a beautiful woman dressed as a man.''

To deal with this, a method has been proposed in which a thesaurus containing synonyms is prepared, the words requested by the user are expanded into synonyms, and the synonyms are matched with the text.

Note that, instead of using a keyword search, a method of performing a high-speed search using words similar to words in an input sentence ('I! in the same sentence) as a search key is, for example, disclosed in Japanese Patent Application No. 1-54460.
Examples include the techniques described in the specifications and drawings of No. 1 and Japanese Patent Application No. 1-76012. These methods analyze the grammatical structure of a sentence and search for only texts that are closer to the expression of the input sentence based on connective relationships, including case relationships between words.

[Problem to be solved by the invention]

In the above-mentioned conventional technology, the Seisho of foreign words is generally a collection of words with the same or similar meanings (for example, the New Dictionary of Lunar Synonyms (Tsukiso Shoten) or the Publication of Classified Words (National Institute for Japanese Language and Linguistics)). According to this, ``Aozora'' can be developed into synonyms such as ``Prayer Sky'' and ``Blue Sky.'' Therefore, if a user requests a search for ``Blue Sky'' in a text that includes the expression ``Blue Sky'', this can be searched, but seasonal expressions made up of multiple words will not be registered in the thesaurus. , it was not possible to search for "blue sky" or "deep blue sky" from "blue sky."

An object of the present invention is to improve these problems and provide a text-based search method that can search for various similar expressions in natural language and has high search accuracy.

[Means to solve the problem]

In order to achieve the above object, the text-based search method of the present invention uses a text base that stores documents written in natural language, and a means (input method) for inputting texts (keywords and question sentences) using natural language. a means for extracting words from an input sentence (a word extraction part), and a means for extracting a character string to be used for a search from the analysis result of the word extraction part and generating a search character string to be a sample for a search (a structure In a text-based search system having a means (text search unit) for searching a text base using the search character string (generating unit), the system searches for kanji that have the same meaning as the kanji in the heading and kanji that have similar meanings. Develop a memorized synonym dictionary, select target words from the input text when searching, and use WI.
The ajiten UFJ section develops the word into synonyms of its constituent kanji and adds the obtained character string to the above-mentioned search character string.

[Operation J] In the present invention, (a) a text base composed of natural language documents is searched, (b) text to be searched is specified in natural language or words (keywords), and (c) input is performed. Search for words used in a sentence by expanding them into kanji that have similar meanings to the kanji that make up the word.

In particular, since it has a synonym dictionary, when reinforcing the generated search structure, it can also be reinforced so that expressions having synonymous meanings with words in the search structure can also be searched.

This makes it possible to search for expressions that could not be searched in conventional type-a dictionaries, that is, expressions that have similar meanings to the words used in the input sentence, making it possible to further improve search accuracy.

〔Example〕

An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is an explanatory diagram showing the functions of a text-based search system according to an embodiment of the present invention, FIG. 2 is an explanatory diagram of a synonym dictionary according to an embodiment of the present invention, and FIG. FIG. 4 is an explanatory diagram of the search structure generated by the structure generation unit in the embodiment. FIG. 4 is an explanatory diagram of the search structure reinforced by the Wn syllabary expansion unit in the embodiment of the present invention.

In FIG. 1, 1 is an input section, and 2 is a word extraction section.

3 is a word dictionary, 4 is a structure generation unit, 5 is a synonym expansion unit, 6
7 is a synonym dictionary, 7 is a text search unit, and 8 is a text base.

This word dictionary 3 is a memory file, and its content is a row and a half of lit words (independent words (in the case of conjugated words, their stems)). Furthermore, as in, for example, Japanese Patent Application No. 1-54460 and Japanese Patent Application No. 1-76012 F3,
To analyze the grammatical structure of a sentence using a new analysis, l
It is necessary to use a dictionary that has the IT word heading, the word's name, and the grammatical information necessary for parsing.

Further, the synonymous academic world 86 stores kanji that express synonymous meanings. Its contents, for example, as shown in Figure 2, consist of a kanji heading and a collection of kanji that have similar meanings to the kanji.

The text base 8 is a collection of sentences to be searched, and is stored in a state that can be directly handled by a computer by some means, such as a magnetic disk or magnetic tape.

Further, the input unit 1 inputs a search request (question) for searching for text in the form of natural language sentences. In other words,
Natural language sentences are imported into the device through character encoding processing such as key operator voice input and character pattern recognition.

Furthermore, the lit Z extraction unit 2 extracts a word for search (independent YB (or its stem in the case of a usage request)) from the text input in the input unit l. In this case, the words are cut out by li
The matching is performed simply from the beginning of the input sentence with the words stored in the word dictionary 3. Note that it is also possible to further improve the degree of closure of extraction by using heuristics such as the longest match method.

Further, the structure generation unit 4 receives the results of the word extraction unit 2, extracts words to be used in the search, and generates a structure (hereinafter referred to as a search structure) to be a sample to be used in the search. Note that in this embodiment, to simplify the explanation, words extracted from the input sentence are ANDed. For example, if "blue sky and white clouds" is input as a question, the words "blue sky" and "white clouds" are ANDed as shown in FIG.

Further, the synonym dictionary opening unit 5 reinforces the search structure so that expressions having a synonymous meaning with the word in the search structure generated by the structure generation unit 4 can be searched for.

Further, the synonym dictionary 6 stores kanji representing similar meanings, and these are referred to in the synonym dictionary opening section 5.

Community-wise, for each J11wI containing Kanji in the M1 funicular structure, lit 3. The synonyms of each kanji that make up ``t'' are written. For example, a search structure augmented with respect to the search structure shown in FIG. 3 is shown in FIG. In this case, the synonyms of the respective kanji characters "Blue", "Sora J", "White", and "R-Cloud" constituting the words "Blue Sky" and "White Cloud" are set as an OR relationship, and their AND is taken. In addition, if F blue sky and white cloud J are input as a question and OR of 'l' various 'blue sky' and 'white gJ is taken, (((blue OR dark blue OR
...) AND (Sky OR Heaven OR...)) OR ((White O
Rf blue OR...) AND (cloud OR...))) is reinforced.

Further, the search section 7 scans the text base 8 using the search structure generated up to the synonym dictionary opening section 5 and outputs one match as a search result.

Next, we will describe the processing procedure when the input sentence "Blue Sky and White Clouds" is sent to subsequent processing in the text-based search system of this embodiment.

FIG. 5 is a flowchart showing the processing procedure of the text-based search system in one embodiment of the present invention.

In this embodiment, when "Blue Sky and White Clouds" is input from the input unit 1 shown in FIG. "Blue sky" and "white clouds" are extracted (502).

Next, the structure generation unit 4 generates a search structure as shown in FIG. 3 (503).

Next, by referring to the synonym dictionary 6, the synonym dictionary opening section 5 expands the word containing the kanji in the search structure into synonyms of the kanji constituting the word, and compares the synonyms. As an OR relationship, the search structure is reinforced as shown in Figure 4 (
504).

Furthermore, according to this search structure, the text base 8
is scanned and those that match the search structure conditions are output (505).

Therefore, for the input sentence "blue sky and white clouds", it is possible to search for expressions such as "white clouds in the blue sky" or "white clouds in the deep blue sky" that cannot be searched in conventional synonym dictionaries.

In addition, in this embodiment, in order to simplify the explanation, when extracting an independent word from an input sentence, a simple term 1! We have described the process of cutting out words in F3, but for example,
As proposed in No. 54460 and Japanese Patent Application No. 1-76012, the grammatical structure of a single sentence is analyzed in more detail, and only texts that are closer to the expression of the input sentence are retrieved based on relationships including case relationships between words. It is also possible. For this purpose, instead of the processing of the single story extraction unit 2 shown in FIG.
Uses a word dictionary that stores word parts of speech and grammatical information.

It is necessary to perform morphological analysis to identify and decompose each word that makes up an alternating current, and syntactic analysis to determine the structure of a sentence from the way these words are connected.

In addition to thesaurus expansion, it is also conceivable to add thesaurus expansion to a word using a thesaurus dictionary containing synonyms that have similar meanings to the word.

[Effects of the Invention] According to the present invention, there is no need to perform prior processing on a text base, it is possible to avoid missing information due to prior processing, and it is easy to use, and words with similar meanings to those used in the input sentence can be avoided. It is possible to search for expressions.

Thereby, search accuracy can be further improved.

[Brief explanation of the drawing]

FIG. 1 is an explanatory diagram showing the functions of a text-based search system according to an embodiment of the present invention, FIG. 2 is an explanatory diagram of a synonym dictionary according to an embodiment of the present invention, and FIG. 3 is an explanatory diagram showing an embodiment of the present invention. An explanatory diagram of the search structure generated by the structure generation unit in the example,
FIG. 4 is an explanatory diagram of the search structure reinforced by the synonym expansion unit in one embodiment of the present invention, and FIG. ff55 is a flowchart showing the processing procedure of the text-based search system in one embodiment of the present invention. 1: Input unit, 2: JQ word extraction unit, 3: Word dictionary. 4: Structure generation unit, 5: Thesaurus expansion unit, 6: Thesaurus dictionary,
7: Text search section, 8: Text base. Figure 3 Concave Figure 3 (Blue sky A, Ni) White clouds) Figure 4 ((Blue OR dark blue OR Ao 01 (101t-) AND (Sky O
Rten OR uit...)ANl,) (white OR$
'Blue oit... )ANI) (Cloud on...
・) Figure

Claims (1)

[Claims] (1) A text base that stores documents written in natural language, a means for inputting sentences using natural language, a means for extracting words from the input text, and the analysis results of the word extraction means are used for searching. In a text-based search system that includes means for extracting a character string and generating a search string to be used as a search sample, and means for searching the text base using the search string, It has a thesaurus dictionary that stores kanji with similar meanings and kanji with similar meanings, selects the target word from the input sentence at the time of search, and expands the word into synonyms of the kanji that constitute the word. , a text-based search method characterized in that the obtained character string is added to the above-mentioned search character string.