JPH03122769A - Keyword associative retrieving device - Google Patents

Abstract

PURPOSE:To reflect a user's own request concept to a retrieving condition expression by calculating the degree of keyword relation indicating the strength of relation of respective keywords included in a keyword dictionary at the time of inputting at lease one retrieving keyword and its weight value. CONSTITUTION:The internal constitution of an associative retrieving part 2 at the time of defining that the whole number of keywords is (n) consists of an input means 5, a combining means 6 and an output means 7. In the case of forming a retrieving condition for information retrieval, a keyword associated from the keyword dictionary by the retrieving keyword is also displayed together with the keyword relating degree. Thereby a retrieving person can search and select the necessary keyword without referring a thesaurus or the like. Consequently, the retrieving person can reflect his (or her) own request concept to the retrieving condition expression.

Description

[Detailed description of the invention] Industrial applications The present invention relates to a keyword associative search device.

Conventional technology Conventionally, this type of system includes, for example, ``Software Knowledge Collection rR System SOCKS (1)J'' and ``Software Knowledge Collection IR System 5OCKS (2)J (both of which are published in the Information Processing Society of Japan's 35-cap National Conference Proceedings). This method uses a knowledge base that holds keyword association rules and opening speed calculation rules to perform related keyword searches and associative searches from keywords input by the user.

Problems to be Solved by the Invention However, the rules that make up the knowledge base, such as the strength of the relationship between synonyms and the strength of the keywords and lessons, are set by several experienced people, and such a knowledge base The creation method has problems with the consistency of the rules as a whole and the relative balance of the strength of relationships. In addition, when creating search conditions for a search, you must use a thesaurus (keyword collection) to find and enter the necessary search keywords, but adding and updating keyword information with a thesaurus is extremely complicated. It is. Therefore, the system does not reflect the user's sense of quantity in the keyword dictionary.

Means for Solving the Problem It has input means for the number of keywords included in a keyword dictionary, output means for the number of keywords, and a weighted coupling coefficient for coupling between the input means and the output means. When at least one search keyword and the weight value of the keyword are input, the combination means calculates the degree of keyword relevance indicating the strength of association with each keyword included in the keyword dictionary, and combines the keyword with the keyword. I made it possible to display the output.

In particular, in the invention set forth in claim 2, in the system set forth in claim 1, the learning control section receives input of a teacher value for a predetermined keyword and executes a predetermined calculation formula, and By inputting a desirable teaching value for a keyword selected from among the keywords, the coupling coefficient is changed by calculation by a learning control unit that reduces the difference between the keyword relevance degree of this keyword and the teaching value.

In addition, in the invention described in claim 3, in the system described in claim 1, a value obtained by normalizing the co-occurrence frequency value of a plurality of keywords in a standard document by the total frequency of appearance of the keyword is calculated based on the relationship between each keyword. , and each coupling coefficient is initialized using the strength of this relationship as an initial value.

When creating search conditions to search for effect information, the keywords associated with the keyword dictionary are also displayed along with their keyword relevance, allowing searchers to search for the necessary keywords without referring to a thesaurus, etc. You can select and reflect your concept of requirements in the search condition expression.

In this case, according to the invention described in claim 3, the co-occurrence frequency value of a plurality of keywords included in the standard document is normalized by the total frequency of occurrence of this keyword, and the value is used as the strength of the relationship between the keywords. It is sufficient to construct a knowledge base called a relational dictionary, which eliminates the problem of setting it up by a specific experienced person.

Further, according to the invention as claimed in claim 2, if the user inputs a desirable teaching value for a keyword that is determined to be related to the search keyword based on the display output and is selected,
The coupling coefficient is changed through the calculation process of the learning control unit, and the strength of the relationship between the keywords can be changed as appropriate based on the judgment of the search user. In other words, the user-adaptive learning function allows the associative ability to be easily modified to suit the user, and the user's sense of various quantities can be directly reflected in the keyword-related dictionary.

Therefore, it can be effectively utilized in a document retrieval device or word group classification device using a keyword-related dictionary.

Example An embodiment of the present invention will be described based on the drawings.

First, FIG. 2 shows the system configuration of a keyword associative search device. This system includes an input analysis section 1 that analyzes search conditions input by a user and transfers a keyword weight value Qk to the associative search section described below, and an input analysis section 1 that associates related keywords from the keyword weight value Qk and searches for each keyword j. It has an associative search unit 2 that calculates keyword accuracy Dj, and an output control unit 3 that sorts all keywords into keyword accuracy classes, generates and outputs a keyword accuracy class list.

Further, the DJ results obtained from the output control section 3 are sorted and displayed to the user in descending order. On this occasion.

The user judges the output results based on the input search conditions,
Desirable value (=teacher value) T as input to input analysis unit l
When j is input, the learning control section 4 changes the coupling coefficient of the coupling means in the associative search section 2 using a predetermined calculation formula.

Here, FIG. 1 shows the internal structure of the associative search unit 2 when the total number of keywords is n, and is composed of an input means 5, a combination means 6, and an output means 7.

The input means 5 consists of n pieces, and the weight value Qk (k=1 to n) is sent to an appropriate combination means 6 according to the selected keyword k.
It is to be distributed to Further, the coupling means 6 distributes to the output means 7 a product Qk*Wkj obtained by multiplying the weight value Qk passed from the input means 5 by a coupling coefficient Wkj. After calculating the sum of such products, the output means 7 performs predetermined threshold processing and outputs the keyword accuracy Dj (j=1 to n).

In such a configuration, an example of its operation will be explained. First, when a user inputs at least one search keyword and each weight value among n keywords as a search condition, the input analysis section 1 inputs the input means 5 corresponding to each keyword number k of the associative search section 2. Input the weight value Qk into . In the associative search unit 2, the weight value Qk is multiplied by a preset coupling coefficient Wkj in the coupling means 6 and sent to the output means 7. The output means 7 calculates the sum of the products calculated by the combination means 6 for all the keywords connected by the combination means 6, and performs threshold processing f on the result. The keyword accuracy Dj is obtained through the above processing.

Here, the keyword accuracy DJ is given by the following equation (1).

Dj= f (net j) ・・・・・・・・・
・・・・・・・・・・・・・・・・・・(1) However, netj=ΣWkj −Qk f(X)=1/(1+e−”).

On the other hand, when the user determines the search condition he/she wishes to enter by referring to the output results and inputs a desirable teacher value Tj as an input to the input analysis section 1, the coupling coefficient Wkj is changed according to equation (2).

Wkj=g(Wkj ten α・ΔWkj) ・・・・・・
(2) where α is a learning coefficient, g is a piecewise linear function, and Wkj with '°'' indicates a value after change.

Here, when the evaluation function E is defined using the teacher value Tj and the current accuracy DJ, it becomes as shown in equation (3).

However, Ej=(Tj-Dj)'/2.

Here, Wkj is changed so that the evaluation function E approaches O by the steepest descent method. Therefore, the change ΔWkj is ``77(Tj Dj)・f' (net j)・Q
kHere, f' (x)-f (x) ・(1-f
(x)), ΔWkj=η(Tj −Dj)・Dj・
(1-Dj)・Qk・・・・・・・・・・・・・・・・
...(4).

In this way, the user can change the coupling coefficient Wkj and learn to improve the keyword association ability.

Effects of the Invention As described above, the present invention comprises input means and output means for the number of keywords included in the keyword dictionary, and a coupling means having a weighted coupling coefficient for coupling between these, and at least one When two search keywords and their weight values are input, the degree of keyword relevance, which indicates the strength of the relationship to each keyword included in the keyword dictionary, is calculated and displayed along with the keywords. When creating search conditions for a search, the keywords associated with the keyword dictionary are also displayed along with their keyword relevance, allowing the searcher to find and select the required keywords without referring to a thesaurus etc. In this case, according to the invention as claimed in claim 3, the co-occurrence frequency value of a plurality of keywords included in a standard document can be calculated based on this keyword. It is sufficient to construct a knowledge base called a keyword relationship dictionary using the value normalized by the total frequency of occurrence as the strength of the relationship between keywords, which solves the problem of setting by a specific experienced person. According to the invention described in 2, if the user inputs a desired teaching value for a keyword that is determined to be related to the search keyword based on the display output and is selected, the coupling coefficient is changed through the calculation process of the learning control unit. The strength of the relationship between keywords can be changed as appropriate based on the judgment of the search user, and with this user-adaptive learning function, the associative ability can be simplified to adapt to the user. It can be changed and the user's word preferences can be directly reflected in the keyword-related dictionary.

[Brief explanation of drawings]

The drawings show an embodiment of the present invention; FIG. 1 is a block diagram showing the internal configuration of an associative search section, and FIG. 2 is a block diagram showing the system configuration. 5...Input means, 6...Coupling means, 7...Output means 1.1 BI◇ ◇ ◇ Applicant Rico Co., Ltd.

Claims (1)

[Claims] 1. Input means for the number of keywords included in the keyword dictionary, output means for the number of keywords, and a weighted coupling coefficient for coupling between the input means and the output means. When at least one search keyword and the weight value of that keyword are input, the keyword relevance degree indicating the strength of association with each keyword included in the keyword dictionary is calculated. A keyword associative search device characterized in that the keyword associative search device is configured to display and output the keywords. 2. It has a learning control unit that receives the input of a teacher value for a predetermined keyword and executes a predetermined calculation formula, and the learning control unit receives the input of a teacher value for a predetermined keyword and executes a predetermined calculation formula. 2. The keyword associative search device according to claim 1, wherein said combination coefficient is changed by calculation by said learning control unit that reduces a difference between a keyword relevance degree of a keyword and said teacher value. 3. The strength of association between each keyword is determined by normalizing the co-occurrence frequency value of multiple keywords in a standard document by the total frequency of occurrence of that keyword, and this value of association strength is used as the initial value for each 2. The keyword associative search device according to claim 1, wherein a combination coefficient is initialized.