JP2013020431A - Polysemic word extraction system, polysemic word extraction method and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a polysemic word extraction system for removing ambiguity of text by identifying a polysemic word, which is used in a meaning different from a general meaning, in a document group of a specific project such as a proposal or specifications for information system construction etc.SOLUTION: The polysemic word extraction system includes: a word analyzing section that extracts words from a given input text; a key word co-occurrence vector extraction section that selects a desired word as a key word and extracts key word co-occurrence words which are in a co-occurrence relation with the key word and a key word co-occurrence vector represented with the number of co-occurrences thereof; a co-occurrence word concept estimation section that estimates a co-occurrence word concept of each key word co-occurrence word of the key word co-occurrence vector on the basis of a general concept; a co-occurrence word classification section that performs clustering of key word co-occurrence words with respect to the selected key word based on the similarity among the corresponding co-occurrence word concepts on the estimated co-occurrence word concept group; a polysemic word candidate estimation section that extracts a polysemic word candidate when plural clusters are found,; and a polysemic word candidate output section that outputs the extracted candidate.

Description

  The present invention relates to a polysemy extraction system, a polysemy extraction method, and a program, and in particular, extracts a polysemy that is assigned a plurality of meanings in a document related to a predetermined matter such as a proposal or a specification for information system construction. The present invention relates to a polysemy extraction system, method, and program.

  In recent years, a system has been developed that uses an information processing apparatus to analyze a document written in a natural language and automatically extract the meaning and significance of the document. Among them, the handling of ambiguous words in sentences may become a problem.

An example of a technique related to a polysemy extraction system is described in Patent Document 1 as a “word thesaurus construction system”. The word thesaurus construction system disclosed in Patent Document 1 includes a sentence analysis unit, an internoun distance calculation unit, a noun clustering unit, an ambiguity elimination unit, a reclustering unit, a thesaurus generation unit, and a data storage unit. The word thesaurus construction system having such a configuration operates as follows.
That is, the sentence analysis unit executes morphological analysis and syntactic analysis of sentences in the corpus to be analyzed, generates verb relation basic data, and generates a noun list, verb list, and co-occurrence relation data. The internoun distance calculation unit obtains the internoun distance based on each generated list and co-occurrence relation data. The noun clustering unit generates a noun cluster based on the calculated internoun distance. The ambiguity elimination unit eliminates the ambiguity of the verb having a co-occurrence relationship with each noun based on the tree structure of the noun cluster, and corrects the verb list and the co-occurrence relationship data. The re-clustering unit performs noun clustering again based on the verb list and the co-occurrence relation data corrected by the ambiguity eliminating unit. The thesaurus generator generates a word thesaurus based on the reclustering result. The data storage unit is a corpus that is a large amount of sentences to be analyzed, basic verb case relationship data generated by analyzing the corpus, a verb list that stores verbs that appear in the sentences together with their appearance frequencies, and , A noun list for storing the nouns appearing in the document together with their appearance frequency, co-occurrence relation data for storing the co-occurrence relations of verbs and nouns in each list, and the distance between nouns calculated by the internoun distance calculation unit And a noun cluster generated by the clustering process and a noun and verb thesaurus generated by the thesaurus generation process. With such a configuration, the ambiguity of the verb is determined for the words in the document based on the distance between the verb and the noun, and the word list and the co-occurrence relation data are corrected based on this determination. It is said that a highly accurate thesaurus can be constructed by clustering nouns again based on this.

Furthermore, another example of the technique related to the polysemy extraction system is described in Patent Document 2 as “Machine Translation Device”. The machine translation device disclosed in Patent Document 2 includes an input unit, an input character string storage unit, a translation dictionary unit, a dictionary search unit, a translation processing unit, a knowledge base unit, a word thesaurus unit, an ambiguity elimination unit, and a translation result output. It consists of parts. The machine translation apparatus having such a configuration operates as follows.
The input unit inputs a source language character string. Next, the input character string storage unit stores the input character string. The translation dictionary section holds source language words and partner language form information, source language and partner language parallel translation information, and the like. The dictionary search unit searches the translation dictionary. When the translation processing unit translates the source language into another language with reference to the translation dictionary unit and identifies the ambiguity in the translation process, the translation processing unit instructs the ambiguity resolution unit to cancel the ambiguity. The knowledge base unit collects co-occurrence relationships between words in the source language and corresponding counterpart language expressions. The word thesaurus section stores semantically similar words. The ambiguity resolution unit eliminates the ambiguity that occurs when the input character string is translated into the partner language. Therefore, the ambiguity resolution unit detects a translation word by referring to the knowledge base, and when it cannot detect, it searches the knowledge base with the original sentence replaced with the semantically similar word in the word thesaurus part, detects a translation word, When it cannot be detected, the translated word is determined by frequency. The translation result output unit outputs the translation processing result. With such a configuration, when ambiguity occurs in a translated word, the scale of the knowledge base is supplemented by a word thesaurus, and the ambiguity is resolved based on an equivalently large knowledge base.

JP 2001-331515 A JP 05-158970 A

  The above-mentioned technical problems can be solved by extracting multiple meanings using the exemplified techniques in the extraction of multiple meanings assigned multiple meanings in documents related to a given project such as proposals and specifications related to information system construction. If is applied, the extraction rate of ambiguous words may be lowered.

  The reason for this is that many of the documents that use such polysemy are generally limited in the amount of sentences, so it is unlikely that the same word will appear as a co-occurrence word for an arbitrary word. This is because it is difficult to perform clustering of co-occurrence words as used in the method of Patent Document 1 on the assumption of the corpus of In other words, the technique of Patent Document 1 has a problem that even if co-occurrence words are clustered based on a small amount of corpus, it is not possible to extract a multiple meaning word with a desired accuracy.

  Another problem is the extraction of multiple meanings using the exemplified technology in the extraction of multiple meanings that are assigned multiple meanings in documents related to a given project, such as proposals and specifications related to information system construction. When the method is applied, it is impossible to extract a polysemy consisting of a document group related to a specific case.

  The reason for this is that it is difficult to grasp the synonym relations in advance for such ambiguous words, and it is necessary to translate them for ambiguousness using the translation dictionary used in the technique of Patent Document 2. This is because it is difficult to determine the presence or absence of a location. For this reason, it is necessary to take measures such as preparing a thesaurus separately from existing dictionaries for special polysemy. However, a great burden is required to prepare the thesaurus.

  Then, it makes it a subject to extract the polysemy required from the document containing the unique polysemy used in the specific range.

  In view of the above, the object of the present invention is to provide a multiple meaning extraction system, method, and method for extracting multiple meanings assigned with a plurality of meanings in a document related to a predetermined matter such as a proposal or specification regarding information system construction. To provide a program.

  The system for extracting multiple meanings according to the present invention includes a word analysis unit that extracts each word used in a predetermined sentence that has been input, selects an arbitrary word from the words as a base word, and the base A basic word co-occurrence vector extraction unit that extracts a basic word co-occurrence vector represented by a basic word co-occurrence word and the number of co-occurrence considered as a co-occurrence relationship with the word, and each basic word co-occurrence vector of the basic word co-occurrence vector The co-occurrence word concept estimation unit that estimates the co-occurrence word concept of the word from the general concept, and the estimated co-occurrence word concept group, each of the selected key word based on the similarity between the corresponding co-occurrence word concepts A co-occurrence word classifying unit that performs clustering of key word co-occurrence words, a polysemy candidate estimating unit that extracts the base word as a multi-word candidate when a plurality of clusters exist for the selected key word, and the extracted polysemy Output many word candidates Characterized in that it comprises a word candidate output unit.

  ADVANTAGE OF THE INVENTION According to this invention, the multiple meaning word extraction system, method, and program which extract the multiple meaning word assigned with several meanings in the document regarding a predetermined | prescribed matter, such as a proposal regarding an information system construction, a specification, etc., can be provided.

It is a block diagram which shows the structure of the polysemy extraction system which concerns on 1st Embodiment. It is a flowchart which shows the operation example of the multiple meaning word extraction system shown in FIG. It is a block diagram which shows the structure of the polysemy extraction system which concerns on 2nd Embodiment. It is a flowchart which shows the operation example of the multiple meaning word extraction system shown in FIG. It is a block diagram which shows the structure of the polysemy extraction system which concerns on a 1st Example. It is explanatory drawing which shows the example of the table | surface which put together the basic word co-occurrence vector Ni. It is explanatory drawing which shows the example of a classification | category system of the general concept information Cg of the thesaurus regarding the basic word co-occurrence word Vij. It is the example which represented co-occurrence word conceptual diagram Cvwj regarding basic word "material" as a tree diagram. It is explanatory drawing which shows the example of the peripheral word structure table VV. It is explanatory drawing which shows the example of the classification system of the general concept information Cg of the thesaurus regarding the peripheral word Vvwjf. It is explanatory drawing which shows the large-category co-occurrence word concept table VC1 based on the peripheral word Vvwjf of the co-occurrence word of the basic word “material”. It is explanatory drawing which shows the co-occurrence word concept table | surface VC2 of the middle classification based on the peripheral word Vvwjf of the co-occurrence word of basic word "material". It is explanatory drawing which shows the co-occurrence word concept table | surface VC3 of the small classification based on the peripheral word Vvwjf of the co-occurrence word of basic word "material". It is explanatory drawing which shows an example of the clustering result based on the dendrogram of co-occurrence word conceptual diagram Cvwj regarding basic word "material". It is explanatory drawing which shows an example of the clustering result based on the dendrogram of co-occurrence word conceptual diagram Cvwj regarding basic word "material". It is a block diagram which shows the structure of the polysemy extraction system which concerns on a 2nd Example. It is explanatory drawing which shows the example of the partial coincidence compound word co-occurrence table VUx containing composition word "process." It is explanatory drawing which shows the example of the partial coincidence compound word co-occurrence table VUx containing a constituent word "change." It is explanatory drawing which shows the example of the compound word structure distribution table Te regarding compound word "change process". It is explanatory drawing which shows the example of the table | surface which put together the basic word co-occurrence vector Ni which considered the compound word. It is explanatory drawing which shows the example of a classification | category system of the thesaurus general concept information Cg regarding the basic word co-occurrence word Vij which considered the compound word. It is the example which represented co-occurrence word conceptual diagram Cvwj which considered the compound word as a tree diagram regarding basic word "material". It is explanatory drawing which shows an example of the clustering result based on the tree diagram of co-occurrence word conceptual diagram Cvwj which considered the compound word regarding basic word "material".

[Embodiment 1]
First, a first embodiment of the present invention will be described in detail with reference to the drawings.
FIG. 1 is a block diagram showing a configuration of a polysemy extraction system 100 according to the first embodiment of the present invention.

Referring to FIG. 1, a multiple meaning extraction system 100 according to the first embodiment of the present invention is basically a system comprising an electronic device or a server and an electronic device, and an information communication network such as the Internet for interconnecting them. Among them, at least the document input unit 10, the word analysis unit 20, the basic word co-occurrence vector extraction unit 30, the co-occurrence word concept estimation unit 40, the co-occurrence word classification unit 50, the multiple meaning word candidate estimation unit 60, and the multiple meaning word candidate output Unit 70 and concept database 110.
The illustrated multiple meaning word extraction system 100 is a multiple meaning word extraction system that extracts a multiple meaning word assigned a plurality of meanings in a document related to a predetermined item such as a proposal or specification regarding information system construction.

When a polysemy extraction system is comprised with an electronic device, the polysemy extraction system 100 is realizable with the computer which operate | moves by program control. Although not shown, this type of computer, as is well known, includes an input device for inputting data, a data processing device, an output device for outputting processing results in the data processing device, and an auxiliary memory serving as various databases. Device. Then, the data processing device stores data in a read-only memory (ROM) that stores a program, a random access memory (RAM) that is used as a work area that temporarily stores data, and a program stored in the ROM. It consists of a central processing unit (CPU) that processes stored data.
In this case, the data processing device functions as the document input unit 10, the word analysis unit 20, the basic word co-occurrence vector extraction unit 30, the co-occurrence word concept estimation unit 40, the co-occurrence word classification unit 50, and the multiple meaning word candidate estimation unit 60. The auxiliary storage device operates as the concept database 110, and the output device functions as the polysemy candidate output unit 70.

  Next, the operation of each component constituting the polysemy extraction system 100 will be described.

  The document input unit 10 receives an input of a document or a document group from which an ambiguous word is extracted.

  The word analysis unit 20 applies morphological analysis or syntactic analysis to each sentence constituting a document or a group of documents, so that nouns, verbs, adjectives, adjective verbs and the like used in each sentence are independent. Words are extracted as words, and word information such as part-of-speech for each word, type of particle used immediately after, and dependency relation between words is extracted as necessary. Note that morphemes may be used as they are instead of independent words.

  The base word co-occurrence vector extraction unit 30 sequentially selects any word used in each sentence extracted by the word analysis unit 20 as a base word, and uses any word information or the like for each base word. A base word co-occurrence vector represented by a base word co-occurrence word that is regarded as a co-occurrence relationship with the base word in the co-occurrence determination rule and the number of the co-occurrence are extracted. Here, as the basic word co-occurrence determination rule, a word that has a dependency relationship with the basic word is used as a co-occurrence word, or a specific particle is used in the same sentence as the basic word. A rule that considers a word as a co-occurrence word can be considered. The number of co-occurrence may be the number of co-occurrence, but may be a frequency obtained by dividing the number of co-occurrence by the total number of co-occurrence words for each basic word. Further, with respect to the basic word co-occurrence word and the number of co-occurrence words, weighting may be performed on a predetermined document as an extraction source based on importance, accuracy, parent-child relationship between documents, and the like.

  The concept database 110 accumulates general concept information such as concept classification and general synonyms, synonyms, usages, etc. of collected words, and general concepts related to the meaning and usage of words in response to inquiries about specific words. A database that retrieves and responds to information. The concept database 110 corresponds to a thesaurus that organizes and organizes words according to the upper / lower relationship, partial / whole relationship, synonym relationship, synonym relationship, and the like of words. A database on the Internet may be used as the concept database 110.

  The co-occurrence word concept estimation unit 40 uses the general concept information in the concept database 110 to estimate the co-occurrence word concept of each basic word co-occurrence word of the basic word co-occurrence vector based on a predetermined concept estimation method.

  As the concept estimation method, the basic word is directly inquired of the concept database 110 for general concept information about each basic word co-occurrence word, and all basic word co-occurrence words of a specific basic word are replaced with the general concept based on the general concept information. A method of using a co-occurrence concept vector as a co-occurrence word concept is preferable. When different basic word co-occurrence words become the same general concept by replacement with the concept, the respective basic word co-occurrence words are merged, and the sum of the co-occurrence numbers is registered in the corresponding location. Further, when a thesaurus in which concepts at a plurality of levels such as major classification, middle classification, and minor classification are registered as general concept information is used as the concept database 110, a basic word co-occurrence concept vector is created for each hierarchy. When different co-occurrence words become the same concept in the basic word co-occurrence concept vector in a broad concept such as classification, the respective co-occurrence words are merged and the sum of the co-occurrence numbers is registered in the corresponding location. In addition, when a synonym dictionary in which synonyms including synonyms are registered as general concept information is used as the concept database 110, the key word co-occurrence words are converted into the corresponding synonyms of the corresponding synonyms and the co-occurrence of the synonyms The co-occurrence number of the corresponding base word co-occurrence word is assigned as a number, and the total number of co-occurrence numbers for each synonym converted with respect to the base word co-occurrence word of the same base word is calculated as the base word co-occurrence concept vector good. If there is no concept corresponding to the basic word co-occurrence word in the concept database 110, the co-occurrence word is not converted to the concept, and the word of the co-occurrence word is treated as a concept as it is.

In addition, as another example of the concept estimation method, the peripheral word composition vector based on the peripheral words existing in the vicinity of the basic word co-occurrence word and the number of existing peripheral words based on the arbitrary peripheral word determination rule for the basic word co-occurrence word A peripheral word composition table summarizing the word co-occurrence words is created, the general concept information is inquired to the concept database 110 for each of the peripheral words in the peripheral word composition vector of the peripheral word composition table, and the peripheral word composition within an arbitrary range Peripheral concept vectors created by converting each peripheral word of each peripheral word composition vector in the table into general concepts for each corresponding basic word co-occurrence word, and corresponding peripheral concepts corresponding to all basic word co-occurrence words of a specific basic word A basic word co-occurrence concept table in which vectors are collected may be used as a co-occurrence word concept.
Here, as the peripheral word determination rule, a range to be regarded as a peripheral according to the characteristics of the document, such as one sentence, all sentences within one paragraph, all sentences within the same item on the table of contents, or the whole document, may be set. The range regarded as the periphery may be changed for each part of speech such as a verb coexisting in one sentence and a noun in a sentence in the same item on the table of contents. Further, whether or not a word has a dependency relationship between words may be used as the peripheral word determination rule. The existence number may be the existence number, or may be a frequency obtained by dividing the existence number by the total number of peripheral words for each basic word co-occurrence word. The peripheral word composition table is a matrix in which each row corresponds to each basic word co-occurrence word and each column corresponds to each peripheral word, and the number of peripheral words for the basic word co-occurrence word is registered as each value of the table. It has been done. When different peripheral words become the same concept in the conversion to the concept, the respective peripheral words are merged and the sum of the number of existences is registered in the corresponding part. Further, when a thesaurus in which concepts at a plurality of levels such as major classification, middle classification, and minor classification are registered as general concept information is used as the concept database 110, a basic word co-occurrence concept table is created for each hierarchy. When different peripheral words become the same concept in the basic word co-occurrence concept table in a broad concept such as classification, the respective peripheral words are merged and the sum of the number of existences is registered in the corresponding location. In addition, when a synonym dictionary in which synonyms including synonyms are registered as general concept information is used as the concept database 110, peripheral words are converted into the corresponding synonyms of the corresponding synonyms and corresponding as the number of existing synonyms. Assigns the number of neighboring words, calculates the total number of co-occurrence for each synonym converted for the neighboring words of the same basic word co-occurrence word as a peripheral word concept vector, and all the basic word co-occurrence words of a specific basic word A basic word co-occurrence concept table may be created by collecting peripheral concept vectors corresponding to. If there is no concept corresponding to the peripheral word in the concept database 110, the co-occurrence word is not converted into a concept, and the word of the co-occurrence word is left as a temporary concept.

  The co-occurrence word classifying unit 50 calculates the similarity between corresponding co-occurrence word concepts for each co-occurrence word concept of the basic word co-occurrence word related to the specific basic word by a predetermined similarity index, and the co-occurrence word Based on the similarity index between concepts, each key word co-occurrence word is clustered. Here, the “similarity index” for determining similarity may be a standard for determining the semantic similarity between co-occurrence word concepts. For example, the co-occurrence word concept includes all basic axes of a specific basic word. A basic word co-occurrence concept vector in which a word co-occurrence word is replaced with a general concept based on general concept information, and when the thesaurus is used as the concept database 110, classification until all basic word co-occurrence words are regarded as the same general concept When the synonym dictionary is used as the concept database 110, the basic word co-occurrence concept vector created by the total number of co-occurrence numbers for each synonym converted with respect to the basic word co-occurrence word is used. A function value that is in a monotonically decreasing relationship with a distance such as a cosine distance or a Euclidean distance is suitable as the similarity index. Further, when the thesaurus is used as the concept database 110 in the basic word co-occurrence concept table in which the co-occurrence word concept is a summary of the peripheral word concept vectors corresponding to all the basic word co-occurrence words of a specific basic word, Calculates cosine distance and Euclidean distance between the neighboring word concept vectors corresponding to each basic word co-occurrence word every time, and the weighting distance and monotonously decrease so that the distance in the deeper classification such as small classification is more important A function value in the relationship is appropriate as a similarity index. The clustering method may be a general method, and hierarchical clustering using dendrograms may be applied, or a virtual periphery where similarity between neighboring words and a monotonically decreasing index are derived as distances. Non-hierarchical clustering such as the k-means method or the Fussy-c-means method may be applied to the word position information.

  The multiple-sense word candidate estimation unit 60 has, for each word as a base word, a plurality of clusters in which the size of the cluster is equal to or greater than an arbitrarily determined threshold from the clustering result of each base word co-occurrence word for each base word. The key word to be extracted is extracted as a candidate for a multiple word that is assumed to be a multiple word, assuming that a plurality of usages are semantically seen. In addition, as the scale of the cluster, the number of co-occurrence of each basic word co-occurrence word belonging to the cluster may be used as an index.

The multiple word candidate output unit 70 outputs the multiple word candidate extracted by the multiple word candidate estimation unit 60. Here, the output form may be output in a required form, and an appropriate form is to output the entire document by clearly indicating the key word of the multiple meaning word in the document by color coding or bold emphasis. In addition, the output form may be a form such as a table from which combinations of multiple meaning words are extracted. Also, as other output forms, the main word is a base word that is a multiple word candidate, each cluster based on the concept of the base word co-occurrence word is an intermediate node, and the base word co-occurrence word belonging to each cluster is an end node For example, a graph in which relationships are linked by links may be displayed, and links with a large number of co-occurrence numbers may be color-coded and emphasized. In addition, as an output form, a quantitative ambiguity is added between the ambiguities with the similarity index used when extracting the ambiguity candidates, and only the ambiguities where the ambiguity is larger than the arbitrarily set threshold. The display may be limited. Or as an output form, depending on the degree of ambiguity between the ambiguity candidates, color coding, emphasis by bold letters, or the size of the characters of the words in the graph may be given. Further, each output form may be selected so that the display form as a base can be shifted to a table or a graph as necessary. Moreover, you may make it selectively output a verb, a noun, etc. as needed.
Next, with reference to the sequence shown in FIG. 1 and FIG. 2, the overall operation of the polysemy extraction system 100 according to the first embodiment will be described in detail. Note that the flowchart shown in FIG. 2 and the following description are examples of processing, and the processing order and the like may be changed or the processing may be returned or repeated depending on the desired effect.

The document input unit 10 receives an input of a target document or document group (step A1 in FIG. 2).
The word analysis unit 20 applies morphological analysis or syntactic analysis to each sentence constituting a document or a group of documents, so that nouns, verbs, adjectives, adjective verbs and the like used in each sentence are independent. Words are extracted as words, and word information such as part-of-speech for each word, type of particle used immediately after, and dependency relation between words is extracted (step A2).

  The basic word co-occurrence vector extraction unit 30 selects an arbitrary word used in each sentence extracted by the word analysis unit 20 as a basic word, and based on word information for each basic word, a predetermined basic word co-occurrence A base word co-occurrence vector represented by a base word co-occurrence word that is regarded as a co-occurrence relationship with the base word in the determination rule and the number of the co-occurrence are extracted (step A3).

  The concept database 110 collects and accumulates general concepts related to the meaning and usage of words based on the concept classification of words and general conceptual information such as general synonyms, synonyms, and usages. The conceptual information is retrieved and responded (step A4).

  The co-occurrence word concept estimation unit 40 estimates the individual co-occurrence word concept for each basic word co-occurrence word of the basic word co-occurrence vector based on a predetermined concept estimation method using the general concept information of the concept database 110. (Step A5).

  The co-occurrence word classifying unit 50 refers to each estimated co-occurrence word concept for each basic word co-occurrence word related to a specific basic word, and determines the similarity between the corresponding co-occurrence word concepts as a predetermined similarity. The basic word co-occurrence words are clustered based on the similarity index between the co-occurrence word concepts (step A6).

  From the clustering result of each basic word co-occurrence word related to a specific basic word, the polysemy candidate estimation unit 60 semantically selects a plurality of basic words that have a plurality of clusters whose size of each cluster is equal to or greater than a predetermined threshold. The candidate is sequentially extracted as a candidate for a multiple meaning word that can be used and a possibility of a multiple meaning word is assumed (step A7).

  The multiple word candidate output unit 70 outputs the multiple word candidates that can be extracted by the multiple word candidate estimation unit 60 (step A8).

Next, the effect of the polysemy extraction system 100 according to the first embodiment of the present invention will be described.
In the first embodiment, a result of clustering co-occurrence words that are semantically similar but do not match as a word by converting a basic word co-occurrence word in a document or document group into a co-occurrence word concept Because it is configured to extract the ambiguous word candidates based on the basic word, even if the number of occurrences of each basic word co-occurrence word is small and the distance between the basic word co-occurrence words tends to be zero, It is possible to grasp whether or not there are a plurality of usage patterns, and it is possible to accurately extract a polysemy assigned a plurality of meanings in a document related to a predetermined case.
Note that the polysemy extraction system 100 according to the first embodiment can be realized as a polysemy extraction method. In addition, the polysemy extraction system 100 according to the first embodiment may be executed by a computer using a polysemy extraction program.

[Embodiment 2]
Next, a second embodiment will be described in detail with reference to the drawings.
FIG. 3 is a block diagram showing a configuration of a multiple meaning word extraction system 100A according to the third embodiment.

  Referring to FIG. 3, the polysemy extraction system 100 </ b> A according to the second embodiment further includes a constituent word dominance calculation unit 35 and a compound word constituent distribution estimation unit 36, and a word analysis unit as will be described later. Except that the operation of the co-occurrence word concept estimation unit is different from that of the multiple meaning word extraction system 100 according to the first embodiment shown in FIG. Therefore, the reference symbol of 20A is attached to the word analysis unit, and the reference symbol of 40A is attached to the co-occurrence word concept estimation unit.

When the illustrated multiple meaning word extraction system 100A is realized by the above-described computer, the data processing apparatus includes a document input unit 10, a word analysis unit 20A, a basic word co-occurrence vector extraction unit 30, a constituent word dominance calculation unit 35, a compound word It functions as the composition distribution estimation unit 36, the co-occurrence word concept estimation unit 40A, the co-occurrence word classification unit 50, and the multiple meaning word candidate estimation unit 60, the auxiliary storage device operates as the concept database 110, and the output device is the multiple meaning word candidate output unit 70. Work as.
Then, the word analysis unit 20A acquires a compound word and a component word of the compound word in the word in the document, and a component word dominance calculation unit 35 calculates a component word dominance for each component word of the compound word, The compound word composition distribution estimation unit 36 creates a compound word composition distribution table in which the concept for each component word of the compound word is weighted based on the component word dominance degree, and the co-occurrence word concept estimation unit 40A performs the basic word co-occurrence Before converting a word to a concept, the co-occurrence number of the basic word co-occurrence word that is a compound word in the basic word co-occurrence vector is converted to the co-occurrence number distributed based on the compound word composition distribution table. .

  Next, the operation of each component constituting the polysemy extraction system 100A will be described.

In addition to the operation of the word analysis unit 20 shown in FIG. 1, the word analysis unit 20A inquires the concept database 110 for general concept information of each extracted word, is not registered in the concept database 110, and has two or more characters Is different from the word analysis unit 20 shown in FIG. 1 in that it is extracted as a compound word. Further, the word analysis unit 20A inquires the concept database 110 for general concept information for every partial character string constituting the compound word, extracts a partial character string in which the general concept information is registered as a significant component word of the compound word, and extracts it. When the significant constituent word is separated from the original compound word, if a partial character string for which general concept information is not registered remains in the concept database 110, the word analyzing unit 20 shown in FIG. And different.
Of the partial character strings constituting a compound word, when there are a plurality of combination patterns of partial character strings for which general concept information is registered in the concept database 110, an optimal combination pattern is selected based on an arbitrary constituent word separation rule. Judgment is made, and significant constituent words and unknown constituent words in the combination pattern are extracted. Here, as a constituent word separation rule, a rule that prioritizes a pattern that minimizes the number of characters of unknown constituent words, a rule that prioritizes significant constituent words that frequently appear as a single word in the input document, Rules that prioritize significant constituent words that frequently appear as single words in the document and rules that combine these are effective. In addition, if a character string that is used in common with other compound words included in the input document is used more than the specified frequency, the remaining character strings excluding the character string are significantly composed. Rules that have priority over words may be used.
The general concept information includes a thesaurus classification, a keyword that directly represents the meaning of a word, a set of synonyms, and the like.
In the following description, a simple constituent word includes a significant constituent word and an unknown constituent word.

The constituent word dominance calculation unit 35 determines a word that co-occurs with a compound word in an arbitrary compound word co-occurrence determination rule based on the word and compound word used in each sentence extracted by the word analysis unit 20A. As a co-occurrence word, a compound word co-occurrence word and the number of co-occurrence are extracted for each compound word, and a compound word co-occurrence table is created by collecting them.
Here, the compound word co-occurrence determination rule includes one sentence, all sentences in one paragraph, all sentences in the same item on the table of contents, the whole document, document title, position in the document group, etc. You may choose according to the characteristics. For example, if the part of speech is a verb, co-occurrence within one sentence, and if it is a noun, the range of the document group may be changed for each part of speech, such as co-occurrence within all sentences within the same item on the table of contents. .
The number of co-occurrence may be the number of co-occurrence, but may be a frequency obtained by dividing the number of co-occurrence by the total number of co-occurrence words for each compound word.
Further, when the word information includes a dependency relationship between words, whether or not the word has a dependency relationship may be used as a compound word co-occurrence determination rule.
The compound word co-occurrence table is a matrix in which each row corresponds to each compound word and each column corresponds to each compound word co-occurrence word. It is registered as.

Further, the constituent word dominance calculating unit 35 is based on the constituent words extracted by the compound co-occurrence table and the word analyzing unit 20A, and the compound word co-occurrence of partially matching compound words including the same constituent word from the compound word co-occurrence table. A compound word co-occurrence vector consisting of words is extracted, and a partially matched compound word co-occurrence table is created for each constituent word. Then, the degree of aggregation between the partial coincidence compound words in the co-occurrence vector space obtained from the compound word co-occurrence vector of the partial coincidence compound word co-occurrence table is calculated as the constituent word dominance.
Here, the co-occurrence vector space may be equivalent to each vector, but may be converted into a vector space weighted by the part of speech of the compound word co-occurrence word. In addition, the degree of aggregation between each partially matched compound word may be any calculation method as long as it is an index representing the degree of dispersion between vectors corresponding to each partially matched compound word. For example, a function having a monotonous decrease relationship with an index indicating dispersion generally used in statistics, such as dispersion, standard deviation, and variation coefficient, may be used, and an inverse of dispersion or an inverse of variation coefficient is suitable.

The compound word composition distribution estimation unit 36 calculates a composition word weighting coefficient between the composition words for each compound word with each composition word dominance calculated by the composition word dominance degree calculation unit 35, and combines the composition word weighting coefficients. Create a word structure distribution table.
The compound word concept composition table is a matrix in which each row corresponds to each compound word and each column corresponds to each composition word of the compound word, and corresponding composition word weighting coefficients are registered.
Here, as a method for calculating a constituent word weighting coefficient, a method that uses a normalized value as an index by dividing the constituent word dominance of each constituent word by the sum of the constituent word dominance for each compound word is effective. is there.

In addition to the operation of the co-occurrence word concept estimation unit 40 described above, the co-occurrence word concept estimation unit 40A performs each co-occurrence of the base word co-occurrence vector of the base word co-occurrence table created by the base word co-occurrence vector extraction unit 30. For the base word co-occurrence word that is a compound word in the word, each coefficient is matched with the required estimation method using a coefficient based on the compound word composition distribution table created by the compound word composition distribution estimation unit 36. Estimate co-occurrence word concepts suitable for compound words. As an example, the co-occurrence word concept estimation unit 40A makes each component word constituting each compound word independent as a basic word co-occurrence word, and is based on the compound word component distribution table created by the compound word component distribution estimation unit 36. The predetermined word estimation method by changing the basic word co-occurrence vector using the value calculated by multiplying the co-occurrence number of the basic word co-occurrence word by the constituent word weighting coefficient of each constituent word as the co-occurrence number of each constituent word, Based on the above, the co-occurrence word concept of each basic word co-occurrence word of the basic word co-occurrence vector is estimated.
As a concept estimation method to be used, a co-occurrence word concept co-occurrence concept table that summarizes peripheral concept vectors corresponding to all basic word co-occurrence words of a specific basic word, taking into account peripheral words including compound words, is used. In this case, for the peripheral words that are compound words among the peripheral words of the peripheral word constituent vectors of the basic word co-occurrence table, each constituent word is made independent as a peripheral word, and the composite word constituent distribution estimating unit 36 The peripheral word composition vector may be changed based on the compound word composition distribution table created in step 1 by using the value calculated by multiplying the number of neighboring words by the constituent word weighting coefficient of each constituent word. .

  Other configurations and functions of the document input unit 10, the basic word co-occurrence vector extraction unit 30, the co-occurrence word classification unit 50, the multiple meaning word candidate estimation unit 60, the multiple meaning word candidate output unit 70, and the concept database 110 are as follows. Since they are the same as those of the embodiment, description thereof is omitted.

Next, with reference to the sequence shown in FIG. 3 and FIG. 4, the overall operation of the polysemy extraction system 100A according to the second embodiment will be described. Note that the flowchart shown in FIG. 4 and the following description are examples of processing, and the processing order and the like may be changed or the processing may be returned as in the first embodiment.
Compared to the operation of the first embodiment described above, the operation of the second embodiment described below is different in that the following operation is added.

In addition to the operation (step A2) of the word analysis unit 20 shown in FIG. 1, the word analysis unit 20A inquires the concept database 110 for the general concept information of each extracted word, is not registered in the concept database 110, and has the number of characters Is extracted as a compound word (step B1).
Further, the word analysis unit 20A inquires the concept database 110 for general concept information for every partial character string constituting the compound word, extracts a partial character string in which the general concept information is registered as a significant component word of the compound word, and extracts it. When the significant constituent word is separated from the original compound word, if a partial character string without registration of general concept information remains in the concept database 110, it is extracted as an unknown constituent word (step B2).

Next, the constituent word dominance degree calculation unit 35 is a word that co-occurs with the compound word in the compound word co-occurrence determination rule based on the word information and the compound word of the word used in each sentence extracted by the word analysis unit 20A. As a compound word co-occurrence word, a compound word co-occurrence word and the number of co-occurrence are extracted for each compound word, and a compound word co-occurrence table is created by collecting them (step B3).
Further, the constituent word dominance calculating unit 35 is based on the constituent words extracted by the compound co-occurrence table and the word analyzing unit 20A, and is a compound word co-occurrence word of partially matching compound words including the same constituent word from the compound word co-occurrence table. Extracts compound word co-occurrence vectors consisting of, creates partial match compound word co-occurrence table for each constituent word, and each partially match compound in co-occurrence vector space obtained from compound word co-occurrence vector of partial match compound word co-occurrence table The degree of aggregation between words is calculated as the constituent word dominance (step B4).

  Next, the compound word composition distribution estimation unit 36 calculates a component word weighting coefficient between the component words for each compound word with each component word dominance calculated by the component word dominance degree calculation unit 35, and summarizes the component word weighting coefficients. A compound word composition distribution table is created (step B5).

The co-occurrence word concept estimation unit 40A, in addition to the operation content of the co-occurrence word concept estimation unit 40 shown in FIG. 1 (step A5), the basic axis of the basic word co-occurrence table created by the basic word co-occurrence vector extraction unit 30 For the basic word co-occurrence word that is a compound word among the co-occurrence words of the word co-occurrence vector, each component word is made independent as a basic word co-occurrence word, and the compound word composition allocation estimation unit 36 creates Based on the word composition distribution table, the basic word co-occurrence vector is changed with the value calculated by multiplying the co-occurrence number of the basic word co-occurrence word by the constituent word weighting coefficient of each constituent word as the co-occurrence number of each constituent word, Based on the predetermined concept estimation method, the co-occurrence word concept of each basic word co-occurrence word of the basic word co-occurrence vector is estimated (step A5 ′).
The operation of other steps is the same as the operation in the first embodiment described above, and a description thereof will be omitted.

Next, effects of the above-described operation of the second embodiment will be described.
In the second embodiment, in addition to the effects of the first embodiment, the constituent word dominance degree is calculated for each constituent word for the compound word in the basic word co-occurrence word, and weighting is performed based on the constituent word dominance degree. Convert to a new concept. As a result, it is possible to extract a polysemy candidate in consideration of compound words in which the general concept information is not registered in the thesaurus or the like, which becomes an obstacle to the conversion from the basic word co-occurrence vector to the basic word concept vector. In addition, it is possible to evaluate the similarity between basic word co-occurrence words even in a group of sentences with many unique compound words, and it is possible to more accurately extract a multiple meaning word assigned a plurality of meanings in a document related to a predetermined case.
Note that the multiple meaning extraction system 100A according to the second embodiment may be realized as a multiple meaning extraction method. Moreover, you may make it perform the polysemy extraction system 100A which concerns on the said 2nd Embodiment of this invention with a computer by a polysemy extraction program.

  Next, with reference to FIG. 5, the operation of the polysemy extraction system 100 according to the first embodiment will be described using a specific first example.

The purpose of the first embodiment is as follows.
First, the polysemy extraction system 100 is a document related to a specific item included in a document D including a polysemy that is also used as a meaning indicating a concept different from a general meaning such as a proposal or specification regarding information system construction. Estimate the ambiguous word candidate A that holds only in the group. Then, the ambiguous word extraction system 100 supports the creation of a glossary and the definition of words related to unregistered terms by outputting the estimation result. Further, in the first embodiment, the polysemy extraction system 100 is composed of a document analysis system Y and an Internet server Z as shown in FIG.
The document analysis system Y operates on the PC terminal possessed by the analyst B, and through the input unit and the output unit, inputs of sentences constituting a group of documents for which the analyst B wants to extract the polygram, Realization of candidate A is realized.
The Internet server Z is connected via a communication network to a PC terminal of the analysis person B who has implemented the document analysis system Y. The Internet server Z can search for general concept information Cg related to word concept classification, general polysemy, synonyms, and usage in response to a query of conceptual information such as the meaning of a word from the document analysis system Y. It is a device to make.

The correspondence between FIG. 5 and FIG. 1 will be described.
The document input unit 10, the word analysis unit 20, the basic word co-occurrence vector extraction unit 30, the co-occurrence word concept estimation unit 40, the co-occurrence word classification unit 50, and the polysemy candidate estimation unit 60 It is included in the system Y. The polysemy candidate output unit 70 operates as an output unit of the PC terminal. The concept database 110 is included in the Internet server Z. The document analysis system Y and the Internet server Z provided with such means operate as follows.

  The document analysis system Y receives from the input unit input of a document D that constitutes a document group for which the analyst B wants to estimate a multiple meaning candidate A that is formed only in the document group related to a specific case. Then, the document analysis system Y applies morphological analysis and syntax analysis to each sentence constituting the document D, decomposes it into words constituting the document, and analyzes the part-of-speech of each word and its dependency relationship, And the verb, the adjective, and the adjective verb are extracted as the word W. Of the verbs, the verbs belonging to the sa line modification use are extracted as the verbs by removing the use part and converting them into so-called nouns.

  Further, the document analysis system Y uses the noun as the base word S among the words W included in the document D, and for each base word Si (i = 1, 2,..., N), a specific base word Sw (i = W), verbs, adjectives and adjective verbs that are in a dependency relationship are extracted as basic word co-occurrence words Vwj (j = 1, 2,..., M), and each basic word co-occurrence word for the basic word Sw The number of co-occurrence of Vwj is tabulated as the co-occurrence number Nwj, and a basic word co-occurrence vector Nw is created. For example, from the document D, the words “material”, “general affairs”,..., Etc. as the basic words Sw, and the “co-occurrence words V” “manufacturing”, “stockpiling”, “mining”, “disposal”, “mixing” , “Purchase process”, “Estimate”, “Order”, “Budget”, “Change process”,... Are extracted, and the co-occurrence number Nwj of the basic word co-occurrence word Vwj for each basic word Sw In the case of FIG. 6, the data set of each row in the table of FIG. 6 corresponds to the basic word co-occurrence vector Ni, the data set of a specific basic word Sw corresponds to the basic word co-occurrence vector Nw, and “material” The basic word co-occurrence vector Nw is expressed as {5, 1, 1, 1, 3, 3, 4, 2, 1, 4.

  The Internet server Z accumulates thesaurus general concept information Cg that is organized by classifying words according to general upper / lower relationship, partial / whole relationship, synonym relationship, synonym relationship, and the like. In addition, the Internet server Z also provides a function such as a search engine that extracts information on an arbitrary word, so that it can be used as a general concept classification of words to be inquired according to an inquiry from the document analysis system Y. The classification, middle classification, and minor classification are extracted as general concept information Cg and presented.

  The document analysis system Y extracts the co-occurrence word concept Cvwj for each of the basic word co-occurrence words Vwj of the basic word co-occurrence vector Nw based on the general concept information Cg inquired of the Internet server Z.

  As a method for extracting the co-occurrence word concept Cvwj, a thesaurus stored in the Internet server Z can be obtained by inquiring the Internet server Z about the general concept information Cg of each basic word co-occurrence word Vwj directly. From the classification system of the general concept information Cg, the co-occurrence word concept C1vwj of the large classification, the co-occurrence word concept C2vwj of the middle classification, and the co-occurrence of the small classification as the co-occurrence word concept Cvwj to which each basic word co-occurrence word Vwj belongs An appropriate method is to extract the word concept C3vwj and create a co-occurrence word conceptual diagram Cvwj collected in a tree structure or the like so that the concept co-occurrence number Ncwj in each classification hierarchy can be understood. This method is a concept direct extraction method. According to the concept direct extraction method, for each basic word co-occurrence word Vwj of the co-occurrence number Nwj of the basic word co-occurrence vector Nw related to the basic word Sw “material” in FIG. 6, the co-occurrence word concept C1vwj as shown in FIG. When the co-occurrence word concept C2vwj and the co-occurrence word concept C3vwj are extracted, the co-occurrence word concept diagram Cvwj is represented by a tree diagram as shown in FIG. In FIG. 8, the concept co-occurrence number Ncwj in the hierarchy of each classification of the co-occurrence word conceptual diagram Cvw is calculated as the sum of the co-occurrence numbers Nwj of the basic word co-occurrence words Vwj belonging to lower levels. It should be noted that compound words such as “change process” and “purchase process” for which general concept information Cg is not registered in the Internet server Z are processed while leaving the words of the basic word co-occurrence words as they are as tentative concepts.

In addition, a concept indirect extraction method will be described below as a method for extracting a more advanced co-occurrence word concept Cvwj. In the concept indirect extraction method, for each basic word co-occurrence word Vwj, verbs, adjectives and adjective verbs that are dependent on each basic word co-occurrence word Vwj, and co-occurrence in sentences in the same item on the table of contents Are extracted as peripheral words Vvwjf (f = 1, 2,..., Y), and the number of co-occurrence of each peripheral word Vvwjf with respect to the basic word co-occurrence word Vwj is counted as the existence number Ljf. A peripheral word configuration table VV is created in which the peripheral words Vvwjf for the word co-occurrence word Vwj are summarized in a table format.
A data set in which the number Ljf of each peripheral word Vvwjf with respect to the basic word co-occurrence word Vwj in the peripheral word configuration table VV is collected is referred to as a peripheral word configuration vector Lj. Obtained from the classification system of the thesaurus general concept information Cg stored in the Internet server Z by making an inquiry to the Internet server Z for the general concept information Cg of each peripheral word Vvwjf in the peripheral word composition table VV . Then, the peripheral word concept C1vwjf of the large classification to which each peripheral word Vvwjf belongs, the peripheral word concept C2vwjf of the medium classification, and the peripheral word concept C3vwjf of the small classification are extracted, and the peripheral word Vvwjf in the peripheral word composition table VV is extracted as the peripheral word The concept C1vwjf is converted into a concept C1vwjf, peripheral words Vvwjf having the same concept are collected, the sum of the number of existence Ljf is registered in the corresponding location, and the co-occurrence word concept table VC1 of the broad classification and the peripheral word Vvwjf in the peripheral word composition table VV C2vfwjf is converted into the same concept, peripheral words Vvwjf are grouped together, and the middle class co-occurrence word concept table VC2 in which the sum of the existing numbers Ljf is registered in the corresponding locations, and the peripheral word Vvwjf in the peripheral word configuration table VV are converted into the peripheral word concept C3vfwjf Co-occurrence of sub-categories in which peripheral words Vvwjf with the same concept are collected and the sum of the number of existence Ljf is registered in the corresponding location To create a concept table VC3.
A data set in which the number Lc1jf of each peripheral word concept C1vfwjf for the basic word co-occurrence word Vwj in the large-category co-occurrence word concept table VC1 is called a large-category co-occurrence word concept vector Lc1j. A data set in which the number Lc2jf of each peripheral word concept C2vfwjf with respect to the basic word co-occurrence word Vwj in the word concept table VC2 is called a middle classification basic word concept vector Lc2j, and the basic word co-occurrence word VC3 in the small classification co-occurrence word conceptual table VC3. A data set in which the numbers Lc3jf of the peripheral word concepts C3vfwjf for the word Vwj are collected is called a small classification co-occurrence word concept vector Lc3j.

  Here, the large classification co-occurrence word concept vector Lc1j, the middle classification basic word concept vector Lc2j, and the small classification co-occurrence word concept vector Lc3j correspond to the co-occurrence word concept Cvwj. For example, as shown in FIG. 6, words such as “manufacturing”, “change processing”,... Are extracted from the document D as the basic word co-occurrence word Vwj, and the peripheral words Vvwjf of these basic word co-occurrence words Vwj are extracted. Extract words such as “use”, “operation”, “construction”, “improvement”, “system change”, “mechanism”, “instantaneous”, “short-term”, “operation”, “high-speed processing”, etc. In this case, the peripheral word composition table VV is a table in which the basic word co-occurrence word Vwj is arranged in each row and the peripheral word Vvwjf is arranged in each column and the number of existence Ljf is described as shown in FIG. Further, the data set in the row of the basic word co-occurrence word Vwj in FIG. 9 corresponds to the peripheral word constituent vector Lj, and the peripheral word constituent vector Lj of “manufacturing” is {0, 3, 2, 0, 4, 0, 1 , 0, 3, 0,... Since both the basic word co-occurrence word Vwj and the peripheral word Vvwjf include nouns, the word previously selected as the basic word co-occurrence word Vwj is a peripheral word when another word is the basic word co-occurrence word Vwj. It may be treated as Vvwjf.

  Further, for each peripheral word Vvwjf in the peripheral word composition table VV of FIG. 9, when a peripheral word concept C1vwjf, a peripheral word concept C2vwjf, and a peripheral word concept C3vwjf as shown in FIG. FIG. 11 is a co-occurrence word concept table VC2 of middle classification, FIG. 12 is a co-occurrence word concept table VC2, and FIG. 13 is a co-occurrence word concept table VC3 of a small classification. This is a table in which Cvwjf is arranged. The number of co-occurrence in the co-occurrence word concept table VC1, VC2, VC3 is “use”, “operation”, “construction”, “construction” in the peripheral word Vvwjf, taking the co-occurrence word concept table VC1 of the large classification as an example. Since the peripheral word concept C1vwjf of “improvement” and “operation” is common to “human activity”, the number of existences in these neighboring words Vvwjf is added to the same basic word co-occurrence word “manufacturing” and “8” is present. Lc1jf. Similarly, since the peripheral word concept C1vwjf of “mechanism”, “instantaneous”, and “short term” is common to “abstract” in the peripheral word Vvwjf, the existence number in these peripheral words Vvwjf is used as the basic word co-occurrence word “manufacturing”. “1” obtained by adding together is the existence number Lc1jf. It should be noted that compound words such as “system change” and “high-speed processing” for which general concept information Cg is not registered in the Internet server Z are processed while leaving the co-occurrence word as a temporary concept. From FIG. 11, the major co-occurrence word concept vector Lc1j of the basic word co-occurrence word “manufacturing” is represented as {8, 4, 1, 0,.

  Furthermore, the document analysis system Y calculates the similarity Fw between each basic word co-occurrence word Vwj based on each co-occurrence word concept Cvwj, and summarizes each basic word co-occurrence word Vwj having a similarity Fw larger than an arbitrary threshold. Then, each basic word co-occurrence word Vwj is clustered to extract each basic word co-occurrence word cluster Ewz.

  As an example of a method for calculating the similarity Fwpq between the basic word co-occurrence word Vwp (j = p) and the basic word co-occurrence word Vwq (j = q), each co-occurrence word concept Cvwj was obtained by the concept direct extraction method. In this case, the co-occurrence word concept Cvwp and the co-occurrence word concept Cvwq are quantified by the hierarchy difference between the classification hierarchy in which the classification is the same and the rough classification hierarchy in the classification system. For example, as shown in the example of FIG. 8, the basic word co-occurrence word Vwp “is a thesaurus having a classification system composed of three layers of a large classification (first layer), a middle classification (second layer), and a small classification (third layer). “Manufacturing” and the basic word co-occurrence word Vwq “mining” coincide with each other in the subcategory co-occurrence concept C3vwj “production”, and “2” is the difference index between the first and third layers. Become. Further, if the threshold value of similarity Fw is 1 or more in the example of FIG. 8, the basic word co-occurrence words Vwj are clustered below the middle-class co-occurrence word concept C2vwj, and a dotted line as shown in FIG. The five clusters surrounded by are extracted as basic word co-occurrence word clusters Ewz.

On the other hand, when each co-occurrence word concept Cvwj is obtained by the concept indirect extraction method, the large-category co-occurrence word concept vector Lc1p corresponding to the basic word co-occurrence word Vwp and the large classification co-occurrence word Vwq are corresponded. Calculate cosine distance dc1pq between word concept vector Lc1q, cosine distance dc2pq between middle classification co-occurrence word concept vectors Lc2p and Lc2q, and cosine distance dc3pq between small classification co-occurrence word concept vectors Lc3p and Lc3q Then, the sum of the respective classification weighting coefficients β1, β2, and β3 (β1 <β2 <β3) is calculated as the basic word co-occurrence word distance dwpq by the following equation (1), and the basic word co-occurrence word such as the reciprocal number is calculated. The similarity Fwpq is calculated by a function that is in a monotonically decreasing relationship with the inter-distance dpq. This process is performed for all combinations of basic word co-occurrence words Vij.
dpq = β1 × dc1pq + β2 × dc2pq + β3 × dc3pq (1)

  For example, in the examples of FIGS. 11 to 13, the cosine distance between the basic words “manufacturing” and “accumulation” is dc1pq = 0.26, dc2pq = 0.57, dc3pq = 0.68, and the classification weighting coefficient is β1 = 0. When 009, β2 = 0.09, and β3 = 0.9, the basic word co-occurrence distance dpq = 0.67. Further, as a clustering method, each basic word co-occurrence word Vwj is regarded as an initial cluster, and with respect to the basic word co-occurrence word distance dpq, the clusters having the closest inter-cluster distance are set as new clusters, and further, between all new clusters. Find the distance, repeat the process of joining the two closest and create a new cluster, create a dendrogram by repeating until all the clusters are combined into one cluster, and summarize with any intercluster distance criterion A group of the obtained basic word co-occurrence words Vwj is set as a basic word co-occurrence word cluster Ewz. Based on the information of FIGS. 9 to 13, when the inter-cluster distance reference is set to 5 in the obtained dendrogram, two clusters are extracted as the basic word co-occurrence word cluster Ewz as shown in FIG.

The document analysis system Y uses the basic word co-occurrence word cluster Ewz obtained as a result of the clustering of the basic word co-occurrence words Vwj for the specific basic word Sw, and the basic word co-occurrence belonging to each basic word co-occurrence word cluster Ewz. The sum of the co-occurrence numbers Nwj of the word Vwj is extracted as the cluster scale Nwz. Then, the base word Sw in which a plurality of clusters having a cluster size Nwz equal to or greater than a predetermined threshold value is extracted as a multiple meaning word candidate Aw in which a plurality of usages are seen and a possibility of a multiple meaning is assumed.
In the examples of FIGS. 6 to 9 in which each co-occurrence word concept Cvwj is obtained by the concept direct extraction method, if the threshold is 20%, there are 25 basic word co-occurrence words Vwj, and therefore five or more basic word co-occurrence “Industry” and “Economy” are extracted as the basic word co-occurrence word cluster Ewz to which the word Vwj belongs, and the basic word Sw “material” is determined as the ambiguous word candidate Aw. From the meaning of the basic word co-occurrence word Vwj belonging to the basic word co-occurrence cluster Ewz “Industry” and the meaning of the basic word co-occurrence word Vwj belonging to the basic word co-occurrence word cluster Ewz “Economy”, “material” is There is a high possibility of having two meanings as abbreviations of “material” and “material procurement department”, and such ambiguity can be found. Similarly, in the example of FIG. 15 in which each co-occurrence word concept Cvwj is obtained by the concept indirect extraction method, if the threshold value is 20%, each of the two clusters includes five or more basic word co-occurrence words Vwj. Therefore, the basic word Sw “material” is determined as the ambiguous word candidate Aw.
Further, the document analysis system Y processes the multiple meaning word candidate Aw corresponding to the multiple meaning word candidate Aw in the request document D, such as color coding or bold emphasis, and outputs the processed request document D from the output unit.

Next, with reference to FIG. 10, the operation of the multiple meaning word extraction system 100A according to the second embodiment will be described using a specific second example.
In the second embodiment, it is assumed that the multiple meaning word extraction system 100A uses an Internet server Z ′ as shown in FIG.
The document analysis system Ya operates on the PC terminal possessed by the analyst B, and through the input unit and the output unit, inputs of sentences constituting a group of documents for which the analyst B wants to extract the polygram, Realization of candidate A is realized.

  The Internet server Z ′ is a server that provides an existing thesaurus, and is connected to a PC terminal possessed by an analysis person B who has implemented the document analysis system Ya via a communication network. The Internet server Z ′ enables retrieval of general concept information Cg related to word concept classification, general synonyms and synonyms, and usage in response to a query of word concept information from the document analysis system Ya. Device.

In the second embodiment, in addition to the operation of the first embodiment, the document analysis system Ya further includes a constituent word dominance calculation unit 35 and a compound word constituent distribution estimation unit 36.
That is, the correspondence between FIG. 16 and FIG. 3 is as follows.
The document input unit 10, the word analysis unit 20A, the constituent word dominance calculation unit 35, the compound word composition distribution estimation unit 36, the basic word co-occurrence vector extraction unit 30, the co-occurrence word concept estimation unit 40A, The word classification unit 50 and the multiple meaning word candidate estimation unit 60 are included in the document analysis system Ya. The polysemy candidate output unit 70 operates as an output unit of the PC terminal. The concept database 110 is included in the Internet server Z ′.

The document analysis system Ya including such a configuration adds the following operation to the first embodiment described above.
The document analysis system Ya makes an inquiry to the Internet server Z ′ for the general concept information Cg of each basic word co-occurrence word Vij, so that the general concept of each word Vij is stored in the thesaurus stored in the Internet server Z ′. Search whether the information Cg is registered, and extract a word having two or more characters as a compound word Vme (e = 1, 2,..., H) without registration of the general concept information Cg in the thesaurus. To do. For example, when the word “purchase processing” is not registered in the thesaurus, it is extracted as a compound word because it has two or more characters.

Further, the document analysis system Ya separates the character string of the compound word Vme in every pattern for each compound word Vme, and the general concept information Cg is stored in the thesaurus stored in the Internet server Z ′ for all the separated partial character strings. Search if is registered. Then, the partial character string is processed as a constituent word Pek (k = 1, 2,..., L) of the compound word Vme in a pattern in which the number of characters of the partial character string without registration of general concept information is the smallest. Among the constituent words Pek, the partial character string in which the general concept information Cg is registered is extracted as a significant constituent word Paek, and the partial character string without registration is extracted as an unknown constituent word Pbek for each compound word.
In the example of the compound word “purchase process” in FIG. 6, {“purchase”, “buy process”}, {“purchase”, “process”}, {“purchase process”, “reason”} can be separated. If “buy process” and “purchase process” are not registered in the thesaurus, “buy”, “purchase”, “process”, and “reason” are candidates for significant constituent word Paek, “buy process” ”,“ Purchase ”is a candidate for the unknown constituent word Pbek, but the combination of {“ Purchase ”,“ Process ”}} having the smallest number of characters in the partial character string for which the general concept information Cg is not registered is the compound word“ Purchase As a significant constituent word.

The document analysis system Ya is a noun that co-occurs with the compound word Vme in the sentence of the paragraph specified by the analyst B as a sentence group that refers to the contents of a certain range in the document D such as “function of the information system to be constructed”. , And the verb and adjective related to the compound word Vme, and the adjective verb as s compound word co-occurrence words Umer (r = 1, 2,..., S), and for each compound word Vme, the compound word co-occurrence word Umer and , The co-occurrence number Mer within the range regarded as co-occurrence is extracted, each row is associated with each compound word Vme, each column is associated with each compound word co-occurrence word Umer, and the compound word co-occurrence word Umer with respect to the compound word Vme. A compound word co-occurrence table VUm composed of a sparse matrix in which the co-occurrence number Mer is registered as each value is created.
Further, the document analysis system Ya uses the row components of t compound words Vmx including the same component word Px (x = 1, 2,..., T) for each component word Pek of the compound word co-occurrence table VUm. (Mx1, Mx2, Mx3,..., Mxs) are extracted, each row component is associated with each compound word Vmx, each column is associated with each compound word co-occurrence word Umxr, and compound word co-occurrence with respect to the compound word Vmx A partially matched compound word co-occurrence table VUx composed of a sparse matrix in which the co-occurrence count Mxr of the word Umxr is registered as each value is created.
For example, a table as shown in FIG. 17 is created as the partially matching compound word co-occurrence table including the constituent word “processing”, and FIG. 18 is created as the partially matching compound word co-occurrence table including the constituent word “change”. Further, the document analysis system Ya uses a data string (M1r, M2r, M3r,..., Mtr) for each compound word co-occurrence word Umxr of the partial match compound word co-occurrence table VUx as shown in the following equation (1). Then, the variance σxr is calculated, and the reciprocal of the square root of the average value of the variance σxr of all compound word co-occurrence words Umxr is calculated as the constituent word dominance Gx of the constituent word Px.

The document analysis system Ya calculates a normalized constituent weighting coefficient αek by dividing the value of each constituent word dominance Gxek corresponding to each constituent word Pek for each compound word Vme by the sum of the constituent word dominance degrees Gxek. , Each word is associated with each compound word Vme, each column is associated with each component word Pek, and each component word Pek with respect to the compound word Vme is registered as each value as a compound word composition distribution table Te. Create
For example, with respect to “change processing” and “purchase processing” that are compound words in the basic word co-occurrence words in FIG. 6, the constituent word control degree Gx of the constituent word “processing” is 1.47, and the constituent word “change” 19 is 2.21, and the constituent word dominance Gx of the constituent “purchase” is 3.43, the compound word constituent distribution table Te is as shown in FIG. FIG. 19 shows that when the compound word “change process” is understood as a combination of the constituent word “change” and the constituent word “process”, the constituent word “change” is more important than the constituent word “process”. ing.

  From the viewpoint that the compound word Vme is one of the basic word co-occurrence words Vij, the document analysis system Ya selects the basic word co-occurrence word Vmwe for each of the constituent words Pek of the compound word Vmwe that co-occurs with the specific basic word Sw. As independent. Then, based on the compound word composition distribution table Te, a value obtained by multiplying the co-occurrence number Nwe of the compound word Vmwe by the component word weighting coefficient αek corresponding to each component word Pek is calculated as the co-occurrence number Nwek, so that the basic word co-occurrence The vector Nw is changed. The key word “material” in FIG. 6 will be described in detail. The compound words “change” and “purchase”, which are compound words “process” and “change”, and “purchase” and “process” are key words. As a co-occurrence word, as shown in FIG. 19, the component word weighting coefficient of “change process” is “change = 0.6”, “process = 0.4”, and similarly the configuration of “purchase process” Since the word weighting coefficients are “purchasing = 0.7” and “processing = 0.3”, the weighted co-occurrence number Nwek is “change: 2.4 = 4 × 0.6”, “processing: 2.5 = 4 × 0.4 + 3 × 0.3 ”and“ Purchase: 2.1 = 3 × 0.7 ”. The other basic word co-occurrence words Vw are processed in the same manner, and the basic word co-occurrence vector shown in FIG. 6 is converted into the basic word co-occurrence vector shown in FIG.

  Other operations of the document analysis system Ya are the same as those in the first embodiment. For example, for each basic word co-occurrence word Vwj of the co-occurrence number Nwj of the basic word co-occurrence vector Nw regarding the basic word Sw “material” in FIG. 20, the co-occurrence word concept C1vwj as shown in FIG. When the co-occurrence word concept C2vwj and the co-occurrence word concept C3vwj are extracted, the co-occurrence word concept diagram Cvwj is represented by a tree diagram as shown in FIG. Furthermore, if the threshold value of similarity Fw is 1 or more in the example of FIG. 22, each basic word co-occurrence word Vwj is clustered below the middle-class co-occurrence word concept C2vwj, and a dotted line as shown in FIG. Are extracted as the basic word co-occurrence word cluster Ewz. Even if the threshold of the cluster scale is set to 30% higher than in the first embodiment, two of “industry” and “economy” are extracted as the basic word co-occurrence word cluster Ewz to which the basic word co-occurrence word Vwj belongs, and the basic word Sw “Material” is determined to be an ambiguous word candidate Aw. As described above, unknown compound words that are not registered in the dictionary are taken into account for each constituent word, so that it is possible to accurately estimate a polysemy that takes into account more basic word co-occurrence words.

  As described above, according to the polysemy extraction system of the present invention, a document having a polysemy that is assigned a plurality of meanings in a document related to a predetermined matter such as a proposal or a specification for information system construction, It is possible to grasp the ambiguous words made up of the document from the document or document group used for the analysis. As a result, confusion and failure due to misunderstandings can be reduced during the construction of information systems. The reason for this is that the similarity of co-occurrence words of words is calculated based on the matching level at the concept level, and the co-occurrence words are clustered, so that the same co-occurrence information can be obtained with a limited amount of documents related to a specific project. This is because even if there is no use of words, it is possible to extract words that have a plurality of co-occurrence word groups and are likely to be ambiguous.

  In addition, the specific configuration of the present invention is not limited to the above-described embodiment, and changes within a range not departing from the gist of the present invention are included in the present invention.

  For example, an information processing device that operates as a polysemy extraction system is extracted from a document received from an input unit in order to extract the polysemy as necessary from a document containing polysemy having a usage in a concept different from the general concept. Are extracted from each word used as a sentence, and the basic word from the extracted word group having a co-occurrence relationship with an arbitrary word as a basic word and the basic word from the co-occurrence number The basic word co-occurrence vector is extracted, the co-occurrence word concept of each basic word co-occurrence word included in the basic word co-occurrence vector is estimated individually, and included based on the similarity between the estimated co-occurrence word concepts Each of the basic word co-occurrence words is clustered, and when there are a plurality of clusters related to the basic word selected as the arbitrary word, the process of setting the basic word as a multiple word candidate is repeated and extracted. And it outputs a candidate from the output unit.

At this time, a weight may be given to each document to be analyzed (each sentence group).
For example, a highly accurate document and a document that is not so may be input together with weighting and used as each coefficient.
Also, the creator of the document group (sentence group) to be analyzed or the affiliated organization may be weighted.
Also, the weight may be calculated based on the citation relationship and the number of citations that the document group has.
Moreover, when extracting the polysemy about the translated sentence, you may make it use the general concept of the original language before translation for the general concept to be used.
These pieces of information may be received from the operator, or the sentence may be automatically extracted by performing natural language analysis such as syntax analysis or semantic analysis.
Further, suitable candidates may be extracted or automatically selected by an algorithm such as a concept estimation method to be used by natural language analysis.

According to the present invention, it is possible to support understanding, creation, and correction of documents by excluding ambiguity of documents for various documents exchanged in work such as requirement definition in software and system development. It can be applied to applications related to streamlining system development, such as reducing rework and improving customer satisfaction.
In addition, since multiple terms can be extracted with high accuracy, they can be used in translation systems for translation.

DESCRIPTION OF SYMBOLS 10 Document input part 20, 20A Word analysis part 30 Basic word co-occurrence vector extraction part 35 Composition word dominance calculation part 36 Compound word structure allocation estimation part 40, 40A Co-occurrence word concept estimation part 50 Co-occurrence word classification part 60 Ambiguous word Candidate estimation section 70 Ambiguous word candidate output section 100, 100A Ambiguous word extraction system D Document Y, Ya Document analysis system Z, Z 'Internet server

Claims (18)

A word analysis unit that extracts each word used in a predetermined sentence that has been input;
A base word that selects an arbitrary word from the words as a base word, and extracts a base word co-occurrence vector represented by the base word co-occurrence word and the co-occurrence number considered to be a co-occurrence relationship with the base word A co-occurrence vector extraction unit;
A co-occurrence word concept estimation unit that estimates a co-occurrence word concept of each basic word co-occurrence word of the basic word co-occurrence vector from a general concept;
About the estimated co-occurrence word concept group, based on the similarity between corresponding co-occurrence word concepts, a co-occurrence word classification unit that performs clustering of each of the basic word co-occurrence words related to the selected basic word,
A multiple word candidate estimation unit that extracts the basic word as a multiple word candidate when a plurality of clusters exist with respect to the selected basic word;
A multiple word candidate output unit that outputs the extracted multiple word candidates;
A polysemy extraction system characterized by comprising A document input unit for receiving input of a target document or document group, and
A word analysis unit for extracting each word used in a sentence constituting a document or a document group, and extracting word information about a part of speech and a case for each word, a particle to be combined, and a dependency relation between words;
An arbitrary word is selected as a basic word among the words, and a basic word co-occurrence word that is regarded as a co-occurrence relationship with the basic word in an arbitrary basic word co-occurrence determination rule based on word information for each basic word and its word A basic word co-occurrence vector extracting unit for extracting a basic word co-occurrence vector represented by the co-occurrence number;
Collects and accumulates general concept information that organizes general concepts of words such as word concept classification, synonyms, synonyms, and usage, and collects general concept information related to the meaning and usage of words for inquiries about specific words A conceptual database that searches and responds;
A co-occurrence word concept estimator that estimates the co-occurrence word concept of each basic word co-occurrence word of the basic word co-occurrence vector based on a predetermined concept estimation method using general concept information of the concept database;
For each basic word co-occurrence word for a specific basic word, the similarity between the corresponding co-occurrence word concepts is calculated by a predetermined similarity index, and each basic word co-occurrence word is calculated based on the similarity index between the co-occurrence word concepts. A co-occurrence classifier that clusters words,
A multiple word candidate estimator for extracting, as a multiple word candidate, a basic word in which a plurality of clusters equal to or larger than a threshold that is arbitrarily determined by the size of each cluster of each basic word co-occurrence word for each basic word;
A multiple word candidate output unit that outputs the extracted multiple word candidates;
A polysemy extraction system characterized by comprising A word analysis unit that extracts each word used in a predetermined sentence that has been input and extracts a compound word and its constituent words within each word;
A constituent word dominance calculation unit for calculating a constituent word dominance for each constituent word;
A compound word composition distribution estimator that calculates a composition word weighting coefficient for each compound word using each constituent word dominance degree;
A base word that selects an arbitrary word from the words as a base word, and extracts a base word co-occurrence vector represented by the base word co-occurrence word and the co-occurrence number considered to be a co-occurrence relationship with the base word A co-occurrence vector extraction unit;
For each basic word co-occurrence word that is a compound word in each basic word co-occurrence vector of the basic word co-occurrence table, each constituent word is treated as a basic word co-occurrence word, and the basic word co-occurrence The basic word co-occurrence vector is updated using the value calculated by multiplying the number of word co-occurrence by the word weighting coefficient of each constituent word as the co-occurrence number of each constituent word, and each basic word co-occurrence word of the basic word co-occurrence vector A co-occurrence word concept estimation unit that estimates a co-occurrence word concept from a general concept;
About the estimated co-occurrence word concept group, based on the similarity between corresponding co-occurrence word concepts, a co-occurrence word classification unit that performs clustering of each of the basic word co-occurrence words related to the selected basic word,
A multiple word candidate estimation unit that extracts the basic word as a multiple word candidate when a plurality of clusters exist with respect to the selected basic word;
A multiple word candidate output unit that outputs the extracted multiple word candidates;
A polysemy extraction system characterized by comprising A document input unit for receiving input of a target document or document group, and
Collects and accumulates general concept information that organizes general concepts of words such as word concept classification, synonyms, synonyms, and usage, and collects general concept information related to the meaning and usage of words for inquiries about specific words A conceptual database that searches and responds;
Extract each word used in a sentence that composes a document or group of documents, and extract word information related to part-of-speech and case for each word, particle to be combined, and dependency relation between words. A word with no general concept information registered and two or more characters is extracted as a compound word, and a partial character string with general concept information registered is extracted from any partial character string constituting the compound word. A word analysis unit that extracts as a significant constituent word and extracts a partial character string without registration as an unknown constituent word;
Based on the word information of each word and the compound word, the compound word co-occurrence rule is used as a compound word co-occurrence word, and the compound word co-occurrence word and the number of co-occurrence are extracted for each compound word. By combining these, a compound word co-occurrence table is created, and a compound word co-occurrence vector composed of compound word co-occurrence words of partially matching compound words including the same component word is extracted from the compound word co-occurrence table, and configured Create a partially matched compound word co-occurrence table for each word, and calculate the degree of aggregation between each partially matched compound word in the co-occurrence vector space obtained from the compound word co-occurrence vector of the partially matched compound word co-occurrence table as the constituent word dominance A constituent word dominance degree calculation unit,
Calculating a constituent word weighting coefficient between constituent words for each compound word using each constituent word dominance degree, and creating a composite word constituent distribution table that summarizes the constituent word weighting coefficients;
An arbitrary word is selected as a basic word among the words, and a basic word co-occurrence word that is regarded as a co-occurrence relationship with the basic word in an arbitrary basic word co-occurrence determination rule based on word information for each basic word and its word A basic word co-occurrence vector extracting unit for extracting a basic word co-occurrence vector represented by the co-occurrence number;
For the basic word co-occurrence words that are compound words among the co-occurrence words in the basic word co-occurrence vector of the basic word co-occurrence table, each component word is treated as a basic word co-occurrence word, and the compound word composition distribution table The basic word co-occurrence vector is updated using the value calculated by multiplying the co-occurrence number of the basic word co-occurrence word by the constituent word weighting coefficient of each constituent word as the co-occurrence number of each constituent word. A co-occurrence word concept estimation unit that estimates the co-occurrence word concept of each basic word co-occurrence word of the basic word co-occurrence vector based on a predetermined concept estimation method using concept information;
For each basic word co-occurrence word related to the arbitrary basic word, the similarity between the corresponding co-occurrence word concepts is calculated by a predetermined similarity index, and each basic word is based on the similarity index between the co-occurrence word concepts A co-occurrence word classifier that performs clustering of co-occurrence words,
A multiple word candidate estimator for extracting, as a multiple word candidate, a basic word in which a plurality of clusters equal to or larger than a threshold that is arbitrarily determined by the size of each cluster of each basic word co-occurrence word for each basic word;
A multiple word candidate output unit that outputs the extracted multiple word candidates;
A polysemy extraction system characterized by comprising   The base word co-occurrence determination rule in the base word co-occurrence vector extraction unit is a rule that regards a word having a dependency relationship with the base word as a co-occurrence word, or with a specific particle in the same sentence as the base word The multi-word extraction system according to claim 2 or 4, wherein a rule that regards a used word as a co-occurrence word is used.   The concept database stores a word with a classification system, and is a thesaurus that can be acquired as general concept information about synonymous relationships, synonymous relationships, upper / lower relationships, and partial / whole relationships between words. The multiple meaning word extraction system according to claim 2 or 4.   The concept estimation method of the co-occurrence word concept estimation unit queries the concept database for general concept information related to each basic word co-occurrence word, and replaces all basic word co-occurrence words of a specific basic word with the general concept information concept The word co-occurrence concept vector is a co-occurrence word concept, and the co-occurrence word classification unit performs clustering using the depth of classification until all basic word co-occurrence words are regarded as the same general concept information concept as a similarity index. The multiple meaning word extraction system according to claim 6. The concept estimation method of the co-occurrence word concept estimation unit calculates all peripheral words that exist around the base word co-occurrence word and the number of peripheral words based on the number of presence words based on any peripheral word determination rule for the base word co-occurrence word. Create a peripheral word composition table summarizing basic word co-occurrence words, inquire general concept information from the concept database for each of the peripheral words in the peripheral word composition vector of the peripheral word composition table, and Peripheral word corresponding to all basic word co-occurrence words of a specific basic word is created for each corresponding basic word co-occurrence word by creating a peripheral word concept vector obtained by converting each peripheral word of each peripheral word constituent vector in the composition table into a general concept A basic word co-occurrence concept table that summarizes concept vectors is used as a co-occurrence word concept.
The co-occurrence word classifying unit calculates the distance between the peripheral word concept vectors corresponding to each basic word co-occurrence word for each hierarchy, and the distance weighted so as to emphasize the distance in the more detailed classification and the monotonic decrease The multi-word extraction system according to claim 6, wherein clustering is performed using function values having a relationship of   The range in which the arbitrary peripheral word determination rule in the concept estimation method of the co-occurrence word concept estimation unit regards as a periphery for each part of speech such as a verb that coexists in one sentence and a noun in a sentence in the same item on the table of contents. The system of claim 8, further comprising an algorithm for changing   Range in which the compound word co-occurrence determination rule of the constituent word dominance calculation unit considers co-occurrence for each part of speech, such as a word having a dependency relationship if the part of speech is a verb, or a word in the same paragraph if the part of speech is a noun 10. The multi-word extraction system according to claim 4, further comprising an algorithm that extracts a compound word co-occurrence word and calculates a compound word co-occurrence number under different conditions.   A function in which the degree of aggregation between the partially matched compound words in the constituent word dominance calculating unit is a monotonously decreasing function as an index indicating the degree of dispersion between vectors corresponding to each partially matched compound word The multi-word extraction system according to any one of claims 4 to 10, which is calculated by:   The degree of aggregation between partially matched compound words in the constituent word dominance calculation unit is calculated based on a vector space weighted by the part of speech of a co-occurrence word. The ambiguous word extraction system according to item 1.   The compound word composition distribution estimation unit calculates a normalized weighting coefficient by dividing the component word dominance of each component word of the compound word by the sum of the component word dominance of each compound word. The multi-word extraction system according to any one of claims 4 to 12.   A weighting factor is given for each document or sentence group to be analyzed, and the estimated co-occurrence word concept of each basic word co-occurrence word for an arbitrary basic word is used to estimate from the general concept. 14. The multi-word extraction system according to claim 1, wherein a cluster is formed using a co-occurrence word concept to determine whether the base word is a multi-word word candidate. Extract each word used in the given sentence that received the input,
An arbitrary word is selected as a base word from the extracted words, and a base word co-occurrence vector represented by the base word co-occurrence word and the co-occurrence number regarded as the co-occurrence relation with the base word is extracted. And
Estimating the co-occurrence word concept of each basic word co-occurrence word of the extracted basic word co-occurrence vector from the general concept,
For the estimated co-occurrence word concept group, based on the similarity between the corresponding co-occurrence word concepts, clustering each of the basic word co-occurrence words for the selected basic word,
A polysemy extraction method, wherein when a plurality of clusters exist for the selected base word, the base word is extracted as a multi-word candidate. Extraction of each word used in a document received from an input unit or a sentence constituting a document group, and extraction of word information related to part-of-speech and case for each word, particle to be combined, and dependency relation between words,
An arbitrary word is selected as a basic word among the words, and a basic word co-occurrence word that is regarded as a co-occurrence relationship with the basic word in an arbitrary basic word co-occurrence determination rule based on word information for each basic word and its word Extract the core word co-occurrence vector represented by the number of co-occurrence,
Collect and accumulate general concept information that organizes general concepts of words such as word concept classification, synonyms, synonyms, and usage, and general concept information related to the meaning and usage of words for inquiries about specific words Using the general concept information obtained as a response from the concept database that retrieves and responds, based on a predetermined concept estimation method, the co-occurrence word concept of each basic word co-occurrence word of the basic word co-occurrence vector is estimated,
For each basic word co-occurrence word for a specific basic word, the similarity between the corresponding co-occurrence word concepts is calculated by a predetermined similarity index, and each basic word co-occurrence word is calculated based on the similarity index between the co-occurrence word concepts. Perform clustering of words,
A polysemy extraction method characterized by extracting a base word in which a plurality of clusters having a cluster size of each base word co-occurrence word relating to each base word that are not less than a predetermined threshold are present as a multi-word candidate. The control unit of the information processing device
A word analysis unit that extracts each word used in a predetermined sentence that has been input;
A base word that selects an arbitrary word from the words as a base word, and extracts a base word co-occurrence vector represented by the base word co-occurrence word and the co-occurrence number considered to be a co-occurrence relationship with the base word A co-occurrence vector extraction unit;
A co-occurrence word concept estimation unit that estimates a co-occurrence word concept of each basic word co-occurrence word of the basic word co-occurrence vector from a general concept;
About the estimated co-occurrence word concept group, based on the similarity between corresponding co-occurrence word concepts, a co-occurrence word classification unit that performs clustering of each of the basic word co-occurrence words related to the selected basic word,
A multiple word candidate estimation unit that extracts the basic word as a multiple word candidate when a plurality of clusters exist with respect to the selected basic word;
A program that is operated as a multiple word candidate output unit that outputs the extracted multiple word candidates. The control unit of the information processing device
A document input unit for receiving input of a target document or document group, and
A word analysis unit for extracting each word used in a sentence constituting a document or a document group, and extracting word information about a part of speech and a case for each word, a particle to be combined, and a dependency relation between words;
An arbitrary word is selected as a basic word among the words, and a basic word co-occurrence word that is regarded as a co-occurrence relationship with the basic word in an arbitrary basic word co-occurrence determination rule based on word information for each basic word and its word A basic word co-occurrence vector extracting unit for extracting a basic word co-occurrence vector represented by the co-occurrence number;
Collect and accumulate general concept information that organizes general concepts of words such as word concept classification, synonyms, synonyms, and usage, and general concept information related to the meaning and usage of words for inquiries about specific words Co-occurrence that estimates the co-occurrence word concept of each basic word co-occurrence vector of the basic word co-occurrence vector based on a predetermined concept estimation method using general concept information obtained as a response from the concept database that searches and responds A word concept estimator;
For each basic word co-occurrence word for a specific basic word, the similarity between the corresponding co-occurrence word concepts is calculated by a predetermined similarity index, and each basic word co-occurrence word is calculated based on the similarity index between the co-occurrence word concepts. A co-occurrence classifier that clusters words,
A multiple word candidate estimator for extracting, as a multiple word candidate, a basic word in which a plurality of clusters equal to or larger than a threshold that is arbitrarily determined by the size of each cluster of each basic word co-occurrence word for each basic word;
A program that is operated as a multiple word candidate output unit that outputs the extracted multiple word candidates.

Images