JPH10171821A - Method for presenting retrieval word candidate and device therefor - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a retrieval word candidate presenting method and device for executing efficient retrieval and obtaining a further proper text. SOLUTION: Words are extracted from a text designated by a user to be related with a retrieval request from among texts included in the previous retrieved result, and the group of words which are left as the result of narrowing-down such as the specification of a part of speed is presented to a user, and a retrieval expression is re-constituted only of words designated as those which are judged to be appropriate by the user from among them. Therefore, the candidates of the words being the elements of a new retrieval expression are presented to the user from the text designated by the user, and the retrieval expression is generated from the words designated by the user from among them. Thus, a further proper text can be obtained by this time retrieval.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for searching a text database, that is, a method and apparatus for presenting search word candidates in a text search.

[0002]

2. Description of the Related Art Text search is a technique in which a set of documents is registered in a database, and documents related to a search formula given by a user are extracted from the database. The search expression is not limited to words like "communication", but also "communication AND
A search expression relating to both words of "communication" and "computer" such as "computer" and a retrieval expression relating to any word such as "communication OR computer" are often accepted.

Also, as shown in FIG. 1, a user gives a search formula to perform a search (step (hereinafter, S) 11), displays the search result (S12), and based on this, the user changes the search formula. (S13) and re-search (S14) are often performed. For example, if you first search for the word "communication" as a search expression and the search result is a collection of a few documents from a narrower range than you want,
Change the search formula to "communication 0R calculator" or the like to broaden search conditions.

[0004] In text search, a technique called relevance feedback has been proposed as a technique for automatically changing the search formula to some extent and performing a search again. See “Information Retrieval” (Wil
Prentice Ha by lam B. Frakes and Ricardo Baeza-Yates
ll publication), described in Chapter 11 of this book, but here is a brief description of relevance feedback.

[0005] First, the flow of a search including general relevance feedback is shown in FIG. A search is performed based on a search formula specified by the user (S21), and the search result is displayed (S22). A new search formula is automatically generated from the text specified by the user as appropriate among the texts obtained as search results (S23). The search is performed again based on the new search formula (S24). In this way, until the necessary information is obtained, the automatic generation of the search formula from the specified text and the re-search are repeated.

Next, FIG. 3 shows a flow of a general search expression generation process in matching feedback. First, a morphological analysis is performed on the designated text to perform word division and part-of-speech identification of each word (S31). For example, if a part of the specified text is "... service C is a new service to compete with competitors ...", the result of the morphological analysis for this part is as shown in FIG. For example, "service C" and "ha" are each divided into one word, and their parts of speech are nouns and particles, respectively. Second, valid words are extracted from the designated text (S32). A frequently used method is to extract only a predetermined part of speech from words included in a specified text. In this method, only a noun is often extracted. For example, when a part of a certain text is as described above, "service C", "competition", "each company", and "service" are selected from this part.

[0007] The words extracted so far are often scored based on the frequency of appearance, and narrowed down to only words with high scores. Various methods are conceivable for assigning points. For example, the following methods are available. That is, words that appear less frequently in the entire text database to be searched among words appearing in the specified text are considered to indicate the characteristics of the specified text more strongly, and are given higher scores. . A simple formula for achieving this is: score given to a word = log (1 ÷ the number of texts in the text data pace that include the word). Only those whose score exceeds the threshold are extracted. The appearance frequency is obtained in advance at the time of text registration.

Third, a search formula is constructed (S33).
A simple method is to OR the extracted valid words with OR ,
That is, it is to construct a search formula that only needs to include any of the valid words. From the text examples given above, the following search formula is constructed.

[…] OR Service C OR Competitive OR OR
Service OR … ”In this way, by extracting relevant terms from the specified text and reflecting them in the search formula, a wide range of related texts that could not be obtained in the previous search is searched for next. Now you can get it.

In some cases, the text obtained by the re-search is scored in a more appropriate order, but is omitted here.

[0011]

In the conventional search technique described above, it is possible to change the search condition with reference to the result obtained in the previous search, but there is no guideline on how to change the search condition. However, efficiency is low because trial and error is performed. On the other hand, in the adaptive feedback technology, although narrowing down, such as identification of part of speech, is performed, since the search expression is generated mechanically from words extracted from the specified text, unnecessary words are included in the generated search expression. There was a problem that even when recalculating, even unrelated text was obtained.

An object of the present invention is to provide a method and apparatus for presenting a search term candidate which can execute an efficient search and obtain a more appropriate text.

[0013]

According to the present invention, in order to achieve the above object, words are extracted from a text specified by a user as being relevant to a search request among texts included in a previous search result, A set of words remaining as a result of narrowing down the part of speech and the like is presented to the user, and a search formula is reconstructed from only words that specify those that the user deems appropriate from among them. In addition, when presenting, the words are presented in order from the more characteristic words in the designated text.

According to the present invention, word candidates to be elements of a new search formula are presented to the user from the text specified by the user, and a search formula is generated from the word specified by the user from among them. It is possible to obtain more appropriate text in this search.

[0015]

DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below.

FIG. 5 is a block diagram showing a search word candidate presenting apparatus according to the present invention, and FIG. 6 is a flowchart showing a search word candidate presenting process. The search word candidate presentation device includes a control unit 51, a display unit 52, and a user designation unit 5.
3, a valid word extraction unit 54, a score calculation unit 55, a refinement unit 56, a search formula generation unit 57, and a search unit 58. Hereinafter, the flow of the search word candidate presentation process will be described.

First, search results are displayed on the display unit 52 (S61). The user specifies the text related to the search request from the displayed text of the search result by the user specifying unit 53 to present the search word candidate (S62).

Second, valid words are extracted from the designated text by the valid word extracting unit 54 (S63). As described in the section of the related art, the extraction of the effective word uses the identification of the part of speech (usually only a noun) and the scoring based on the frequency of the word. The score of the effective word is calculated by the score calculation unit 5
5 (S64), and the narrowing-down unit 56 performs narrowing using this (S65). Up to this point, it is the same as the prior art.

Third, a set of words extracted by the above-described methods is presented to the user on the display unit 52 in the order of the highest score as search word candidates (S66). The user specifies a plurality of appropriate words from the presented words (S67). This is a point different from the prior art.

Fourth, a search formula is generated by the search formula generation unit 57 using the word specified by the user (S69), and the search is performed by the search unit 58 (S69).

The processing flow of the present invention will be described more specifically with reference to examples. In response to the displayed search result, as a text associated with the search request, the user has actively promoted the service C in terms of promotion of CF and the like, indicating the strength of Company C's efforts. , The number of contracts for these services has increased significantly, making them dollar boxes. " Nouns extracted from this text by extracting valid words are CF, advertisement, face,
Service C, company A, power, investment, condition, service, contract,
It is a dollar box. Further, a score is assigned based on the same frequency statistic as that described in the description of the related art described above, and narrowing-down is performed by removing those below the threshold value. It is assumed here that the candidates of the additional words finally extracted are arranged in the order of score, such as dollar box, CF, service C, company A, competition, series, and so on. This is presented to the user,
A search expression is generated by connecting the items designated by the user from among them by OR . For example, if the word specified by the user is company A, service C, series, the generated search formula is company A OR service COR series, and the search is performed again using this as a new search condition.

[0022]

As described above, according to the present invention, it is possible to obtain a more appropriate text in this re-search. As a result, the user can more easily obtain the necessary information in a shorter time.

[Brief description of the drawings]

FIG. 1 is a flowchart showing a general flow of “search-change search formula-re-search”;

FIG. 2 is a flowchart showing a flow of a general adaptation feedback;

FIG. 3 is a flowchart showing a general search expression generation process in matching feedback;

FIG. 4 shows an example of morphological analysis.

FIG. 5 is a block diagram of a search word candidate presenting apparatus according to an embodiment of the present invention.

FIG. 6 is a flowchart showing search word candidate presentation processing according to one embodiment of the present invention;

 ──────────────────────────────────────────────────の Continued on the front page (72) Inventor Masayuki Sugizaki 3-19-2 Nishishinjuku, Shinjuku-ku, Tokyo Nippon Telegraph and Telephone Corporation

Claims (2)

[Claims] 1. A method of presenting a search word candidate in a text search, wherein a search result and a search word candidate are displayed, and a user designates a text in the displayed search result and a search word in the search word candidate. , Extract valid search terms from the specified text, calculate the number of search term candidates, narrow down the search term candidates based on the score, generate a search formula from the specified search terms, and generate the generated search formula A method for presenting a candidate for a search word, characterized by performing a search using the search term. 2. An apparatus for presenting a search word candidate in a text search, comprising: a display unit for displaying a search result and a search word candidate; and a user displaying text in the displayed search result and a search word in the search word candidate. , A valid word extraction unit that extracts valid search words from the specified text, a score calculation unit that calculates the number of search word candidates, and a narrowing unit that narrows down search word candidates based on the points A search term candidate presentation device, comprising: a search term generation unit that generates a search term from a specified search term; and a search unit that performs a search using the generated search term.