JPH0498461A - Keyword extracting device - Google Patents

Abstract

PURPOSE:To improve working efficiency in retrieving and the retrieving performance of a device by calculating significance as a keyword based on the number of times of the appearance of a keyword candidate and the constituting word of it and extracting the keyword from among the candidates based on the result of it. CONSTITUTION:A keyword candidate is prepared by a keyword candidate prepar ing means 3 based on a word extracted by a word extracting means 2 and the storage contents of a word information storage means 1 to correspond to the word and siginificance as the keyword of the keyword candidate is calcu lated by a siginificance calculating means 4 based on the number of times of the appearance of the keyword candidate and the constituting word of it. Since the keyword is extracted by a keyword extracting means 5 from among the keyword candidates based on a calculated result, the extracted keyword is not limited to the specified one. Thus, working efficiency in retirieving and the retrieving performance of a device can be improved.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of the Invention The present invention relates to a keyword extraction device for extracting words and phrases that may be used when searching document information as keywords in electronic devices such as database systems and word processors.

Conventional technology Conventionally, when searching for desired information from a large amount of information that has already been created in electronic devices such as database systems and word processors, keywords are assigned to each piece of information in advance and stored. Generally, a conditional expression including a keyword is input at the time of a search, and based on this search condition, information having a keyword matching the search condition is searched and output.

In the case of such electronic devices, the user generally has to register keywords when registering information. Since this registration work is extremely troublesome, it is operationally impossible to assign a sufficient number of keywords to one piece of information in order to facilitate its search.

Furthermore, since at least one keyword must be registered for each piece of information, a large amount (many types) of information cannot be registered in a short period of time.

On the other hand, in order to extract keywords without human intervention, words to be used as keywords are set in advance in a storage means to form a word list, and the words in the target document are set in the word list. There is a keyword extraction method that automatically extracts things as keywords.

In the case of such a keyword extraction method, keywords are extracted from among the very limited words set in the word list, so there are no restrictions on what can be expressed with the words in the word list, i.e. field-specific restrictions, or , there are restrictions on compound words. Furthermore, when registering a new document, it becomes a problem whether or not a keyword corresponding to the document is prepared.

As a solution to this problem, there is a system that automatically extracts keywords without using a word list as described above. A typical example of such a system is a search system called SMART.

This search system includes a morphological analysis means for recognizing words, a list of unused words, a means for functionally canceling high-frequency words from this unused list, and a means for canceling conjugations, prefixes, suffixes, etc. to form the original word. and docum by calculating the importance from the number of occurrences of the original form in - documents.
The keyword extracting means includes a means for creating an entry vector, a means for creating a keyword of a compound word by combining original forms of high importance, and the like.

Problems to be Solved by the Invention In this type of keyword extraction means, even if it is a natural compound word, it is difficult to extract words that include unused words (e.g. research
h and development) cannot be created as a keyword.

On the other hand, many conjugated words in English are irregular.
Even if a method is used to return a word to its original form by canceling a suffix or the like, it is not possible to completely return the target word to its original form.

In addition, there are cases where it is better not to return it to its original form. because,
This is because if only the original form is left, there will be no information on parts of speech, and it will be impossible to see the relationships between modifications and word order (e.g., Japanese book cf, BO).
oks on Japanese).

Furthermore, even if keywords for compound words are created by combining the original forms, when performing a search, the operator does not use compound words based on the original forms as keywords, but instead sets keywords as search conditions using more natural expressions. It is expected that there will be many.

Therefore, when a keyword is set in a natural expression and a search is performed in this way, it may not be possible to search for a target document.

Means for Solving the Problems Word information storage means that stores in advance word information including a large number of predetermined words and their respective parts of speech; word extraction means for extracting words from English information; and the word extraction means. a keyword candidate creation means for creating a keyword candidate based on the word extracted by the second keyword candidate creation means and the stored content of the word information storage means corresponding to the word; and a keyword candidate created by the second keyword candidate creation means and its constituent words. 8, an importance calculation means for calculating the importance of the keyword candidate as a keyword based on the number of current occurrences; and a keyword extraction means for extracting a keyword from the keyword candidates based on the calculation result by the importance calculation means. .

A keyword candidate is created by the keyword candidate creation means based on the word extracted by the action word extraction means and the memory content of the word information storage means corresponding to this word, and based on the number of occurrences of this keyword candidate and its constituent words. The importance of the keyword candidate as a keyword is calculated by the importance calculation means, and based on the calculation result, keywords are extracted from the keyword candidates by the keyword extraction means, so the extracted keywords are limited to specific ones. There is no. Furthermore, keyword candidates can be created based on natural expressions expected to be used during a search, and keywords with high importance as keywords can be extracted from among these keyword candidates.

Embodiment An embodiment of the present invention will be described with reference to FIGS. 1 to 3. As shown in FIG. 1, this keyword extraction device includes a word information storage means 1 that stores in advance a large number of predetermined words and word information corresponding to each word, and a word information storage means 1 for extracting words from English document information. an extraction means 2; a keyword candidate creation means 3 for creating keyword candidates based on the word extracted by the word extraction means 2 and the stored content of the word information storage means 1 corresponding to the word; and this keyword candidate creation means. an importance calculation means 4 for calculating the importance of the keyword candidate as a keyword based on the number of occurrences of the keyword candidate created in step 3 and its constituent words; It consists of a keyword extraction means 5 for extracting keywords from among the keyword candidates created by the creation means 3.

However, the word information stored in the word information storage means 1 consists of part-of-speech information and component part-of-speech information. Furthermore, the part-of-speech information consists of a word and its part of speech, an example of which is shown in Table 1.

Further, the constituent part-of-speech information consists of a part of speech and a position where the part of speech can be a constituent molecule of a keyword candidate, and an example thereof is shown in Table 2.

When extracting keyword candidates in the configuration shown in Table 1 and Table 2, as shown in Figure 2, the keyword candidate generation means extracts keyword candidates from a text file in which English document information is stored. Perform the creation process. That is, first of all, it is determined whether the text file has ended or not, and if it has not ended, words are extracted from this text file by the word extraction means, and the part-of-speech information stored in the word information storage means 1 is extracted. After reading out the part of speech corresponding to this word with reference to , it is determined whether keyword candidates are being created.

If it is determined that keyword candidates are not being created, refer to constituent part-of-speech information stored in the word information storage means 1 (hereinafter omitted) to determine whether the part-of-speech corresponding to the word can be the first word; If it is determined that the word can be at the beginning, creation of keyword candidates is started, and it is further determined whether the part of speech corresponding to the word can be at the end.

If it is determined that the part of speech corresponding to the word can be at the end, after completing the creation of one keyword candidate, creation of the next keyword candidate is started following the end of this keyword candidate.

Then, when you start creating the next keyword candidate, or when it is determined that the part of speech cannot be the first part in the above-mentioned process of determining whether the part of speech can be the first, or whether the part of speech can be the last. If it is determined in the above-described determination process that the part of speech cannot be the last part of speech, the process subsequent to the above-described determination process for determining whether the text file has ended is repeated.

On the other hand, in the above-described process for determining whether keyword candidates are being created, if it is determined that keyword candidates are being created, it is determined whether the part of speech corresponding to the extracted word can be the last word.

If it is determined that the part of speech corresponding to the word can be the last part of speech, the word is continued at the end of the keyword candidate being created, and then the process from the end of creating the keyword candidate described above is repeated, and the part of speech corresponding to the word is continued. If it is determined that this word cannot be placed at the end, it is determined whether this word can be placed in the middle.

If it is determined that the word can be placed in the middle, the word is continued at the end of the keyword candidate being created, and the above-mentioned process for determining whether the text file has ended is repeated until the word is placed in the middle. If it is determined that it is not possible,
After the keyword candidate being created is rejected and the creation of the keyword candidate is completed, the above-described process of determining whether the text file is finished or not is repeated.

Further, in the above-described determination process as to whether the text file has ended, when it is determined that the text file has ended, this keyword candidate creation process is ended.

Here, recently selected as the P
resident ofCzechoslovakia
, Vaclav) label, international
nally recognized as a lead
figure in the world
arts and 1etters, was...

When the above-mentioned keyword candidate generation process is performed on a relatively long phrase, the extracted word, its part of speech, the position where that part of speech can be a constituent molecule of a keyword candidate, and the extent to which processing is performed on the extracted word. Table 3 shows step numbers indicating the number of steps. Furthermore, the keyword candidates created as a result of such processing are
iaaclav Vaclav Havel 1eadingfigure world world of arts iMorld of arts and 1etter
It becomes s.

Also, recent research and devel
opment projects.

Undertaken by Japanese se
Microconductor makers.

Table 3 have...

When the above-mentioned keyword candidate creation process is performed on the phrase , the creation result is research research and development
search and development p
projectsJapanese semicondu
ctorJapanese semiconductor
r makers, and the processing at that time is as shown in Table 4.

Table 4 Next, as shown in FIG. 3, for all of the keyword candidates created in this way, the importance calculation means calculates the importance of each keyword candidate.

That is, each word constituting each keyword candidate, which is either a noun, an adjective, or a verb, is extracted as one word, and the number of occurrences wc of each word is calculated.

Furthermore, the importance S of each keyword candidate is calculated based on the number of appearances wc calculated in this way. The formula used to calculate the importance level S differs depending on the number n of constituent words in each keyword candidate.

In other words, the number of constituent words n for one keyword candidate
is greater than or equal to a predetermined value (limn), the importance S of the keyword candidate is calculated by the formula S-ΣWc11 and depends only on the number of appearances WC of the constituent words.

In addition, when the number of constituent words n for one keyword candidate is smaller than a predetermined value (limn), when the number of appearances of the keyword candidate itself is pc, and the weight to be added to this number of appearances pc is w, The importance level S of the keyword candidate is calculated by the formula S=Σwe +wtXpc, and in addition to the number of appearances WC of the constituent words, the number of appearances of the keyword candidate itself depends on pc.

Further, based on the degree of importance S of the keyword candidates calculated in this way, keyword extraction means extracts keywords from among the keyword candidates.

Since keywords are extracted in this way, the extraction results are not limited to specific ones. Further, keyword candidates are created based on natural expressions that are expected to be used during a search, and furthermore, only keywords with high importance as keywords are extracted from these keyword candidates.

Therefore, when searching for documents, it is possible to easily create appropriate keywords as search conditions, and at the same time, it is possible to improve the work efficiency during the search and the search ability of the device.

Effects of the Invention As described above, the present invention creates a keyword candidate by the keyword candidate creation means based on the word extracted by the word extraction means and the stored content of the word information storage means corresponding to this word, and creates the keyword candidate. The importance calculation means calculates the importance of the keyword candidate as a keyword based on the number of occurrences of the keyword and its constituent words, and the keyword extraction means extracts the keyword from the keyword candidates based on the calculation result. is not limited to specific things.

In addition, keyword candidates can be created based on natural expressions that are expected to be used during searches, and
From these keyword candidates, keywords with high importance can be extracted. Therefore, when searching for documents, it is possible to easily create appropriate keywords as search conditions, and at the same time, it is possible to improve the work efficiency during searching and the search ability of the device.

[Brief explanation of the drawing]

The drawings show an embodiment of the present invention; FIG. 1 is a complaint correspondence diagram, FIG. 2 is a flowchart showing keyword candidate creation processing, and FIG. 3 is a flowchart showing importance calculation processing. l... Word information storage means, 2... Word extraction means, 3
...Keyword candidate creation means, 4.Importance calculation means, 5.Keyword extraction means 1-1 Dou Keizu

Claims (1)

[Claims] word information storage means that stores in advance word information including a large number of predetermined words and their respective parts of speech; word extraction means that extracts words from English text information; keyword candidate creation means for creating keyword candidates based on the memory content of the word information storage means corresponding to the word; and the keyword candidate based on the number of occurrences of the keyword candidate created by the keyword candidate creation means and its constituent words. A keyword extraction device comprising: an importance calculation means for calculating the importance of a keyword as a keyword; and a keyword extraction means for extracting a keyword from the keyword candidates based on the calculation result by the importance calculation means.