JP2529418B2 - Document search device - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to a document search device using a computer.

2. Description of the Related Art In recent years, with the spread of word processors and personal computers and the practical use of character recognition by computers, the number of electronic documents created by these has increased.
For this reason, interest in a document database for accumulating a large amount of document information and searching for the document information as needed is increasing. In a conventional document database, when searching for a document, a keyword search using a keyword attached to each document is general. However, problems such as keyword addition work not keeping up with the number of stored documents, keywords becoming obsolete over time, and being unable to respond to searches with keywords that exceed the expectations of the database administrator, resulting in frequent omission of searches. was there. Against this background, a document database called a full-text database has recently attracted attention. The full-text database collates the search condition given by the user with all the information of the stored documents, and outputs the document satisfying the search condition. In the search condition, a character string such as a sentence may be used in addition to a word such as a conventional keyword, and a notation such as a regular expression can be used as in the conventional case. The matching between the search condition and the document information is performed by using a finite state automaton or the like to determine whether the character strings match or do not match.

Such a full-text database has the following advantages over the document database based on the keyword search. First of all, it is easy to construct a document database because the keyword addition work is unnecessary. Secondly, since all the information in the document is a reference for comparison, the degree of freedom in setting search conditions is high and the omission of search can be reduced.
Thirdly, since the keywords do not become obsolete over time, the document database can be easily maintained.

The problem to be solved by the invention However, in the full-text database, the search condition is set so that the matching succeeds no matter where the collated character string is in the document. Is often successful, and as a result, unnecessary documents are often output as search noise. For example, "arson incident" for the search condition "fire",
For example, "article culture" for the search condition "culture". This type of search noise increases as the length of the matching character string decreases.

Further, the relationship between a plurality of matching character strings that can be expressed by a conventional search condition is a logical sum and a logical product. Therefore, for example, a search condition that represents a modified modification relation between “white” and “dog” such as “white dog” is generally expressed as a logical product of “white” and “dog”. is there. Under such a search condition, even a character string such as "a dog on white snowfield" succeeds in matching, which also becomes a search noise.

In view of the increase in search noise in the full text database as described above, it is an object of the present invention to reduce search noise without increasing search omissions when searching document information.

Means for Solving the Problems In order to achieve the above object, the present invention sets a search condition by a user by inputting a sentence, analyzes the conditional sentence, extracts syntax information, and extracts the syntax information. The document information is searched by using as a search condition.

Operation The operation in the above configuration is as follows. The conditional information for the search condition setting input by the user is parsed and the syntax information is extracted. Next, a morpheme character string suitable for the search condition is selected from the syntactic information of the conditional sentence, a character string representing the same concept as the morpheme is extracted from the synonym dictionary, and the logical sum between the character strings is used using these character strings. Create a search expression by combining and AND. A primary search by character string collation is performed on the document database using this search formula. Next, the sentence in the primary search result is analyzed to extract the syntactic information, the information obtained is compared with the information obtained from the conditional sentence, and the document having the sentence that is successfully collated is output as the final retrieval result.

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

FIG. 1 is a block diagram of a document search device according to an embodiment of the present invention. In FIG. 1, 101 is a search condition input means including a keyword or the like by which a user inputs a search range and a conditional sentence for setting a search condition, 102 is an analysis means for analyzing a sentence to extract syntax information, and 103 is a search Expression generation means, 104
Is a synonym dictionary 105 is a search formula editing means for the user to edit the search formula generated by the search formula generating means 103, and 106 is a character string search for searching a sentence satisfying the search condition of the search formula by character string matching. Means 107, a document database storing document information, 108 a syntax collating means for judging whether the candidate sentence obtained from the character string searching means 106 has the same syntax information as the conditional sentence, 109 a final It is a result display means for displaying the search result and accepting subsequent processing requests.

The syntax information handled by the document search device of this embodiment is 2
It is represented by ternary data consisting of individual clause information and the relation between them, or a set of ternary data. The relations to be handled are the relations of the main part predicates between clauses and the relations of the modified target. The relations other than these two types are not used. Further, the relationship of dependency is represented by associating the clause information of the second term with “dependency” and the clause information of the third term with “acceptance”. BN for syntax information
The F notation (Bacchus-Naur notation) is as shown in FIG.

Next, a specific operation of the document search device according to this embodiment will be described.

The user inputs a sentence or a part of the sentence that is considered to exist in a desired document as a conditional sentence a into the analysis unit 102 via the search condition input unit 101. Further, the search range b is input to the search expression generation means 103 via the search condition input means 101.

The analysis means 102 performs morphological analysis and syntactic analysis on the conditional sentence a, extracts syntactic information c, and retrieves expression generating means.
103 and the syntax collating means 108. When there is a plurality of pieces of syntax information by analysis, the fact is input to the search condition input means 10
The processing is continued by notifying the user via 1 and selecting one piece of syntax information that is considered appropriate from the plurality of pieces of syntax information. If the analysis is impossible, a request for inputting a new conditional sentence is sent to the user via the search condition input means 101.

The search expression generation unit 103 extracts a morpheme suitable for the search condition, for example, an independent word from the syntax information of the conditional sentence a, adds a word having the same concept as the morpheme using the synonym dictionary 104, and then adds the search expression d. To generate. The search expression d is composed of a character string representing a morpheme, a symbol representing a logical sum or logical product relationship between the character strings, and a search range b input via the search condition input means 101. The generated search formula d is presented to the user via the search formula editing means 105, and the user edits the search formula d as necessary. The confirmed search expression d is input to the character string search means 106, and the synonym information e in the search expression is input to the syntax matching means 108.

The character string search means 106 performs character string matching between the content of the document within the specified search range of the document database 107 and the search condition specified by the search expression d, and searches for a sentence that satisfies the search condition. . The search result is the candidate sentence f and the candidate sentence f.
Is a set of pairs of document identifiers g representing the documents to which the belongs. Each candidate sentence f is input to the analysis unit 102 for each sentence, and at the same time, the document identifier g is input to the syntax matching unit 108.

The analysis unit 102 performs morphological analysis and syntactic analysis on the candidate sentence f, extracts syntactic information h, and syntactic collation unit 10
Enter in 8. When there is a plurality of pieces of syntax information at the time of analysis, a predetermined number of pieces of syntax information are input to the syntax collating means 108 in the most probable order.

In the syntax collating means 108, the syntax information c of the conditional sentence a previously input from the analyzing means 102 and the search expression generating means.
The synonym information e input from 103 is collated with the syntax information h of the candidate sentence f input from the analyzing unit 102. If the syntax information of the candidate sentence f satisfies the syntax information of the conditional sentence a, it is assumed that the syntax matching has succeeded, and the document identifier g representing the document to which the candidate sentence f to which the character string searching means 106 belongs is displayed. The document identifier i is input to the means 109. When there are a plurality of pieces of syntax information of the candidate statement f, if any one of the pieces of syntax information satisfies the syntax information of the conditional statement, the candidate statement f and the conditional statement a are collated. Are considered successful.

The result display unit 109 presents the document identifier i from the syntax matching unit 108 to the user. Further, according to an instruction from the user, the content of the document represented by the document identifier i is retrieved from the document database 107 and displayed, and the search condition input means 101 or the search formula editing means 1 for performing a new search.
Transfer control to 05, and perform other processing.

The above is the operation of the document search apparatus of the present embodiment. Further, the operation will be described using an example sentence. For example, when the conditional sentence is "a large white dog runs", the syntax information obtained by the analyzing means 102 is as shown in FIG. In FIG. 3, the morpheme information is a character string representing the morpheme in the sentence. The search condition generated by the search formula generating means 103 is obtained by extracting only an independent word from the syntax information of the conditional sentence. That is, the search condition is "a sentence with {white}, {large}, {dog}, and {running} or {running}". in this case,
{Running} is added as a synonym for {Running}. The character string search means 106 searches the document database 107 for a sentence satisfying this search condition by character string matching. It is assumed that the candidate sentence as shown in FIG. 4 is obtained by the search by the character string search means 106. The syntax matching means 108 performs matching using the relationship between the phrase information and matching using the information about the independent word in the morpheme information forming the phrase. The syntax information of candidate sentence 1, candidate sentence 2, and candidate sentence 3 in FIG. 4 is as shown in FIG. 5, FIG. 6, and FIG. 7, respectively. In FIG. 7,
Two types of syntactic information are obtained. In FIG.
{Size}, {Dog}, and {White} are in the clause information of the conditional sentence shown in FIG. 3, and the relation between the clause information including them is also the same. Also, {run} is a synonym for {run}, and the relationship between the phrase information including {dog} and the phrase information including {run} is {dog} in the syntax information of the conditional sentence.
It is the same as the relation between the phrase information including the phrase and the phrase information including the {run}. Therefore, the candidate sentence 1 and the conditional sentence are
The document identifier of the document to which the candidate sentence 1 belongs is sent to the result display means 109 after the syntactic matching is successful. On the other hand, the syntax information of the candidate sentence 2 and the candidate sentence 3 shown in FIGS.
Either one of the bunsetsu information including {white} and the bunsetsu information including {size} is not related to the bunsetsu information including {dog}, so that the syntax matching fails. In this way, only the candidate sentence 1 is displayed on the result display means 109 as a desired search sentence.

As described above, the present invention uses the syntactic information and synonym dictionary obtained by analyzing the conditional sentence for the search condition setting input by the user, and the logical sum and the logical sum between the character strings. By creating a search formula based on a combination of products and performing a primary search by character string matching on the document database using this search formula, the increase in search omissions is suppressed, and then the sentences in the primary search results are analyzed. Search noise can be reduced by collating the obtained syntactic information with the rear information obtained from the conditional sentence and outputting the document having the sentence successfully collated as the final retrieval result.

[Brief description of drawings]

FIG. 1 is a block diagram of a document retrieval device in an embodiment of the present invention, and FIG. 2 is a BNF of syntax information handled in the embodiment of the present invention.
FIG. 3 is a diagram showing an example of an expression by a notation (Bacchus-Naur notation), FIG. 3 is a diagram showing syntax information of an example of a conditional sentence used in the present invention, and FIG. 4 is an example of candidate sentence used in the present invention. FIG. 5, FIG. 6, FIG. 6 and FIG. 7 are diagrams showing the syntax information of each candidate sentence shown in FIG. 101 …… Search condition input means, 102 …… Analysis means, 103 ……
Search expression generation means, 104 ... Synonym dictionary, 105 ... Search expression editing means, 106 ... Character string search means, 107 ... Document database, 108 ... Syntax matching means, 109 ... Result display means.

Claims (1)

(57) [Claims] 1. A search condition input means for a user to input a search range and a conditional sentence for setting a search condition, an analyzing means for analyzing a sentence to extract syntax information, a synonym dictionary, and the analysis. A sentence having a character string satisfying the search condition of the search expression obtained from the search expression generating means, and a search expression generating means for generating a search expression using the syntactic information of the conditional sentence obtained from the means and the synonym dictionary. A character string search means for searching, a document database in which document information is stored, syntax information obtained by analyzing the sentence obtained by the character string search means by the analysis means, and syntax information of the conditional sentence. It comprises a syntax collating means for collating and outputting a document identifier corresponding to the sentence successfully collated, and a display means for displaying a document corresponding to the document identifier obtained from the syntax collating means in accordance with an instruction from the user. Book search device.