JP5076575B2 - Synonym expression extraction system, synonym expression extraction method, and synonym expression extraction program - Google Patents

Description

  The present invention relates to a synonym expression extraction system, a synonym expression extraction method, and a synonym expression extraction program for extracting synonym expressions, and more particularly, to extract synonym expressions from a small number of similar sentence pairs without specifying special grammar rules in advance. The present invention relates to a synonym expression extraction system, a synonym expression extraction method, and a synonym expression extraction program.

  Synonymous expressions are words or phrases that use different expressions for the same thing or concept. The existence of synonymous expressions in an electronic document is a factor that reduces processing efficiency when performing natural language processing such as document search and document classification. For example, when a document search is performed using “waste material” as a keyword, an electronic document written using the synonymous expression “waste wood” is “waste material” even though it is an electronic document dealing with waste material. Search is not possible because the word is not used.

  Not only does natural language processing become less efficient, but it is created when an electronic document using synonymous expressions is used in an environment where many people are involved, for example, in a product development project. Synonymous expressions in electronic documents can interfere with communication.

  For example, an electronic document created by one person may be expressed as "paraphrase identification technology" in an electronic document created by another person that is the same as that expressed as "synonymous expression extraction function" . In this case, it is difficult for a third party to recognize these “synonymous expression extraction function” and “paraphrase identification technique” as the same thing.

  In order to be able to recognize synonymous expressions used between electronic documents, it is generally desirable to make the synonymous expressions into a dictionary in advance. However, synonymous expressions related to specialized expressions are difficult to deal with such as dictionary. Although it is conceivable that the dictionary is sequentially registered manually, it is burdensome for the user, requires a lot of labor, and is not realistic. Therefore, it is desired that synonymous expressions can be efficiently extracted from the stored electronic document.

  In order to enable automatic extraction of synonymous expressions, sentence pairs having similar contents (hereinafter referred to as similar sentence pairs) are often used. For example, a plurality of articles written on the same news are used as similar sentence pairs. In addition, a plurality of translated books obtained by translating the same book by different translators are used.

  In a certain project, a new electronic document is created by referring to an electronic document created by another member in the project. Therefore, in a project or the like, a group of electronic texts that have a reference relationship with each other also becomes a similar text pair. Such similar sentence pairs often include synonymous expressions related to technical terms.

  In the conventional technology, synonymous expressions are extracted from pairs of similar sentences by statistically processing co-occurrence information between words or using syntactic analysis such as dependency analysis. . The dependency analysis is an analysis process for specifying a modification relationship or a modified relationship between clauses when a sentence is divided into clauses, and an analysis result expressing the sentence as a tree structure by the dependency analysis is obtained. The conventional synonymous expression extraction technique using similar sentence pairs based on syntax analysis will be described below.

  For example, Patent Literature 1 describes an example of a synonymous expression extraction technique. “Information processing apparatus, synonym pair generation method, synonym pair generation program, recording medium recording synonym pair generation program” described in Patent Document 1 is used to generate a synonym database.

  In the synonym expression extraction method described in Patent Document 1, first, a character string 1 in front of a predetermined symbol (for example, “()”) in a certain sentence and a character string 2 in the symbol are extracted. Next, a sentence including the character string 1 or the character string 2 is extracted in addition to the sentence including the extracted character string, and dependency analysis is applied to the extracted sentence. When a predetermined phrase having a relationship between these two character strings is present in the sentence, those character string pairs are extracted as synonymous expressions. Therefore, it can be said that the synonym pair generation method described in Patent Document 1 is a technique for narrowing down synonym expressions from synonym expression candidates using similar sentence pairs.

  Further, for example, Patent Document 2 describes another synonymous expression extraction technique. The “synonym pair extraction apparatus and computer program therefor” described in Patent Document 2 is used to extract synonym pairs from a small amount of data. In the synonym expression extraction method described in Patent Document 2, first, a word string pair including two character strings common to similar sentence pairs and different word strings sandwiched between these two character strings is converted into a synonym expression. Extract as a candidate. Next, from the extracted synonym expression candidates, those satisfying the condition such as not included in the comparison target sentence are extracted as synonym expressions.

  Further, for example, Non-Patent Document 1 describes another synonymous expression extraction technique. The “synonymous expression acquisition method for information extraction” described in Non-Patent Document 1 is used to extract synonymous expressions from a plurality of articles related to the same news. In the synonym expression extraction method described in Non-Patent Document 1, dependency analysis is applied to a pair of similar sentences, and a subtree that satisfies the following three conditions is extracted as a synonym expression. That is, (1) the root of the subtree obtained by dependency analysis is a predicate, (2) the paired subtrees include a common specific expression, and (3) the case required by each predicate. A subtree that satisfies one of the conditions included in the subtree is extracted as a synonym expression. In addition, the proper expression is a numerical expression such as a date or a monetary amount in addition to proper nouns such as a person name, a place name, and an organization name.

  FIG. 32 is an explanatory diagram showing the basic principle of synonymous expression extraction common to the synonymous expression extracting methods described in Patent Document 2 and Non-Patent Document 1. As shown in FIG. 32, in the basic principle used in Patent Document 2 and Non-Patent Document 1, the same marker word (701a, 801a) and synonym candidate (701b, 801b) exist in similar sentence pairs. If the relationship (701c, 801c) between the marker word and the synonym expression candidate is similar, the synonym expression candidate is regarded as a synonym expression. In the synonym expression extraction methods described in Patent Document 2 and Non-Patent Document 1, synonym expressions are extracted from a small number of similar sentence pairs based on such a basic principle.

  The marker word (701a, 801a) is a common character string used in the similar sentence pair in the synonym expression extraction method described in Patent Document 2, and is unique in the synonym expression extraction method described in Non-Patent Document 1. Is an expression. The synonym word candidate (701b, 801b) is a word string sandwiched between common character strings in the synonym expression extraction method described in Patent Document 2, and is the same in the synonym expression extraction method described in Non-Patent Document 1. Is a subtree containing the proper expression. The relationship (701c, 801c) is a relationship in which the synonymous expression extraction method described in Patent Document 2 is sandwiched between common character strings or does not exist in sentences to be compared. In addition, the relationship (701c, 801c) is a relationship that the subtree includes the case required by each predicate in the synonymous expression extraction method described in Non-Patent Document 1.

JP 2006-260402 (paragraphs 0034-0058, FIG. 3) JP 2006-251843 (paragraphs 0022-0048, FIG. 6) Shinyama Y, and Sekine S, "Paraphrase acquisition for information extraction", 2nd International Workshop on Paraphrasing: Paraphrase Acquisition and Applications, pp.65-71, 2003

  As a first problem, the synonym expression extraction methods described in Non-Patent Document 1 and Patent Document 2 cannot extract synonym expressions from pairs of similar sentences that do not satisfy a predetermined grammar rule. That is, in the synonym expression extraction methods described in Non-Patent Document 1 and Patent Document 2, in synonym expression extraction based on a marker word, regularity and grammatical rules of a sequence of similar sentence pairs or a part of character strings thereof are used. It is important that they are similar. For this reason, the synonymous expression extraction methods described in Non-Patent Document 1 and Patent Document 2 impose strong restrictions on the character string arrangement or grammar in advance for sentence pairs from which synonymous expressions are extracted.

  Further, in the synonymous expression extraction method described in Non-Patent Document 1, it is necessary that a similar sentence pair has the same specific expression in a subtree or includes a case required by a predicate. Therefore, synonymous expressions can only be extracted from sentences that meet the grammatical rules. In addition, in order to impose grammatical restrictions, it is necessary to register in advance what kind of case the precaution is required.

  Further, in the synonym expression extraction method described in Patent Document 2, when the word order of the character strings common to the similar sentence pairs is changed, the synonym expression cannot be extracted. Also, since synonymous expressions are not always sandwiched between common words, many synonymous expressions cannot be extracted when synonymous expressions are not sandwiched between common words. Patent Document 2 also mentions the use of syntax analysis, but does not show any specific method. Therefore, even if the parsing is performed, the obtained result is the same as when the parsing is not performed. The reason is that in the synonym expression extraction method described in Patent Document 2, a character string sandwiched between two common character strings is used as a candidate for the synonym expression, and the sandwiched character string does not change even if syntax analysis is performed. Because. Therefore, a synonym expression cannot be extracted unless it is sandwiched between two common character strings.

  As a second problem, the synonymous expression extraction methods described in Non-Patent Document 1 and Patent Document 1 cannot extract synonymous expressions without a special dictionary or symbol expression. The reason is that, as described above, in the synonymous expression extraction method described in Non-Patent Document 1, it is necessary to register in advance the case required by the precaution. A specific expression dictionary is also required. In recent years, the quality of the specific expression dictionary is increasing, but it is difficult for electronic documents exchanged in projects such as reports, specifications and plans to include specific expressions other than numerical information.

  In addition, an organization name or the like is a name that can be communicated only in a specific place such as a department name. Even if a general dictionary is used, the organization name or the like cannot be determined as a unique expression and extracted. Therefore, it is necessary to prepare a special proper expression dictionary. In the synonymous expression extraction method described in Patent Document 1, it is necessary that a symbol expression such as “()” is included in a sentence. However, synonymous expressions are rarely specified by a specific symbol such as “()”. Further, since the technique described in Patent Document 1 is based on the premise that an expression pair that is a candidate for a synonym expression is included in one sentence, synonym expressions are obtained from a state in which only similar sentence pairs are given. It cannot be extracted.

  As a third problem, a method based on statistical processing cannot extract synonymous expressions from a small amount of similar sentence pairs. The reason is that a large number of similar sentence pairs are required for statistical processing to be effective. That is, sufficient accuracy cannot be ensured only by performing processing using only a small number of document pairs. There are a lot of electronic documents in the project, but for example, there are not a lot of pairs of similar sentences that contain synonymous expressions for certain technical terms. That is, synonymous expressions cannot be extracted from a small amount of similar sentence pairs.

  Accordingly, an object of the present invention is to provide a synonym expression extraction system, a synonym expression extraction method, and a synonym expression extraction program capable of extracting a synonym expression from a pair of similar sentences without specifying a grammar rule in advance. To do. The present invention also provides a synonym expression extraction system, a synonym expression extraction method, and a synonym expression extraction program capable of efficiently extracting similar sentence pairs without using special dictionary preparation or symbolic expressions. With the goal. Furthermore, an object of the present invention is to provide a synonym expression extraction system, a synonym expression extraction method, and a synonym expression extraction program that can extract a synonym expression with high accuracy even from a small number of similar sentences.

Synonymous expression extraction system according to the present invention is a synonymous expression extraction system for extracting a synonymous expression from similar sentence pairs paired sentences similar to each other, between the common representation is a representation contained in each sentence similar sentence pairs Based on the structural relationship specified by the structural relationship specifying means and the structural relationship specifying means for specifying the structural relationship between the difference expression and the common expression, which are expressions included only in each sentence of the similar sentence pair The difference expression included in the similar sentence pair is used as a synonym candidate, the similarity calculation means for obtaining the similarity of the synonym expression candidate, and the common expression group included in each sentence of the similar sentence pair in each sentence The relative positional relationship is similar in the sentence structure of the sentence, and the difference expression group and the common expression group that are different between the sentences in the similar sentence pair are similar in the sentence structure of each sentence. Based on Rukoto, synonymous expression extracting means for extracting the synonymous expression from similar sentence pair (e.g., as implemented by the similarity determination unit 104) and a synonymous expression extracting means, the similarity calculation means calculated similarity Based on the degree, it is determined whether or not to extract a synonym expression candidate as a synonym expression .
Another aspect of the synonym expression extraction system according to the present invention is a synonym expression extraction system that extracts a synonym expression from a pair of similar sentences in which sentences that are similar to each other are paired, and the sentence structure that specifies the sentence structure of the pair of similar sentences Based on the sentence structure specified by the specifying means and the sentence structure specifying means, a common expression that is an expression that is commonly included in each sentence of a similar sentence pair, and an expression that is included only in each sentence of the similar sentence pair Based on the expression extraction means for extracting the difference expression, the common expression and the difference expression extracted by the expression extraction means, the relative positional relationship between the common expressions in the sentence structure of each sentence of the similar sentence pair, and the common expression and the difference Based on the relative positional relationship specified by the positional relationship specifying means and the positional relationship specifying means for specifying the relative positional relationship between the expressions, the difference expression in the similar sentence pair based on the identity of the structural relationship The similarity calculation means for calculating the similarity and the common expression group included in common in each sentence of the similar sentence pair have a similar relative positional relationship in the sentence structure of each sentence, and the sentence of the similar sentence pair A synonymous expression extracting means for extracting a synonymous expression from a pair of similar sentences based on the fact that the difference expression group and the common expression group that are different between each other are in a similar relative positional relationship in the sentence structure of each sentence, The synonym expression extraction unit is characterized in that the synonym expression is extracted based on the similarity calculated by the similarity calculation unit.
Still another aspect of the synonym expression extraction system according to the present invention is a synonym expression extraction system that extracts synonym expressions from a pair of similar sentences in which sentences similar to each other are paired, and the sentence that specifies the sentence structure of the pair of similar sentences Based on the sentence structure specified by the structure specifying means and the sentence structure specifying means, a common expression that is commonly included in each sentence of the similar sentence pair and an expression that is included only in each sentence of the similar sentence pair Relative positional relationship between the common expression and the difference expression in the sentence structure of each sentence of the similar sentence pair based on the expression extraction means for extracting a certain difference expression and the common expression and the difference expression extracted by the expression extraction means Based on the relative positional relationship specified by the positional relationship specifying means and the positional relationship specifying means for specifying the similarity, the similarity calculation for calculating the similarity of the difference expression in the similar sentence pair from the identity of the structural relationship A common expression group commonly included in each sentence of a similar sentence pair is in a similar relative positional relationship in the sentence structure of each sentence, and a different expression group that is different between sentences of the similar sentence pair Synonymous expression extraction means for extracting synonymous expressions from pairs of similar sentences based on the fact that the common expression group is in a similar relative positional relationship in the sentence structure of each sentence. Synonymous expressions are extracted based on the similarity calculated by the calculating means.

  Further, in the synonym expression extraction system, the synonym expression extraction means, based on the fact that the difference expression group and the common expression group are similar relative positions in the sentence structure of each sentence in the similar sentence pair. May be extracted.

  Further, the synonym expression extraction system uses a predetermined sentence conversion rule to convert a sentence included in a similar sentence pair (for example, a sentence shaping means) so that the sentence structure of similar sentences has a similar structure. (Which is realized by 105).

  In the synonym expression extraction system, the synonym expression extraction unit extracts the synonym expression from the pair of similar sentences according to a predetermined similarity criterion for extracting the synonym expression, and the synonym expression extraction system extracts the synonym expression extraction unit. Based on correct / incorrect information indicating the correctness of the extraction result of the synonym expression and the similarity of the synonym expression, the similarity reference correction unit (for example, the similarity reference correction unit 107) corrects the similarity criterion for extracting the synonym expression. May be provided.

The synonym expression extraction method according to the present invention is a synonym expression extraction method for extracting synonym expressions from pairs of similar sentences in which sentences similar to each other are paired, and the computer is an expression included in each sentence of the pair of similar sentences. The structural relationship between the common expressions and the structural relationship identifying step for identifying the structural relationship between the common expression and the difference expression that is an expression included only in each sentence of the similar sentence pair, Based on the difference expression included in the similar sentence pair as a candidate for the synonymous expression, the similarity calculation step for obtaining the similarity of the candidate for the synonymous expression, and the common expression included in each sentence of the similar sentence pair Groups are in a similar relative positional relationship in the sentence structure of each sentence, and a difference expression group and a common expression group that differ between sentences of similar sentence pairs are in the sentence structure of each sentence. There Based on that the relative positional relationship similar to, and a synonymous expression extracting the synonymous expression from similar sentence pairs computer, interchangeably expression extraction step, based on the obtained similarity, synonymous expression It is characterized in that it is determined whether or not to extract candidates as synonymous expressions .
Another aspect of the synonym expression extraction method according to the present invention is a synonym expression extraction method for extracting synonym expressions from a pair of similar sentences in which sentences similar to each other are paired, and the computer specifies the sentence structure of the pair of similar sentences. A sentence structure identifying step, a computer based on the identified sentence structure, a common expression that is commonly included in each sentence of a similar sentence pair, and an expression that is included only in each sentence of the similar sentence pair An expression extraction step for extracting a certain difference expression, and a relative positional relationship between the common expressions in the sentence structure of each sentence of the similar sentence pair and a difference from the common expression based on the common expression and the difference expression extracted by the computer Based on the relative positional relationship specified by the computer and the positional relationship specifying step for specifying the relative positional relationship between the expression and the similar sentence from the identity of the structural relationship The similarity calculation step for calculating the similarity of different expressions in the computer, and the computer is in a relative positional relationship in which the common expressions included in each sentence of the similar sentence pair are similar in the sentence structure of each sentence And synonymous expressions that extract synonymous expressions from pairs of similar sentences based on the fact that different expression groups and common expression groups that differ between sentences of similar sentence pairs are in a similar relative positional relationship in the sentence structure of each sentence. Including a representation extraction step, wherein the computer extracts the synonym expression based on the calculated similarity in the synonym expression extraction step.
Still another aspect of the synonym expression extraction method according to the present invention is a synonym expression extraction method for extracting a synonym expression from a pair of similar sentences in which sentences similar to each other are paired. A sentence structure identifying step to be identified, a common expression that is a common expression included in each sentence of a similar sentence pair, and an expression included only in each sentence of the similar sentence pair based on the sentence structure identified by the computer And a relative positional relationship between the common expression and the difference expression in the sentence structure of each sentence of the similar sentence pair based on the common expression and the difference expression extracted by the computer. Based on the relative positional relationship specified by the computer and the positional relationship specifying step, the similarity of the difference expression in the similar sentence pair based on the identity of the structural relationship The similarity calculation step to be calculated, and the computer has a common relative expression group common to each sentence of the similar sentence pair in a similar relative positional relationship in the sentence structure of each sentence, and the sentence of the similar sentence pair A synonymous expression extraction step for extracting a synonymous expression from a pair of similar sentences based on the fact that the difference expression group and the common expression group that are different between each other are in a similar relative positional relationship in the sentence structure of each sentence, The computer is characterized in that the synonym expression is extracted based on the calculated similarity in the synonym expression extraction step.

The synonymous expression extraction method includes a sentence conversion step in which a computer converts a sentence included in a similar sentence pair using a predetermined sentence conversion rule so that the sentence structure of the similar sentence becomes a similar structure. It may be.

In the synonym expression extraction method, the computer extracts synonym expressions from pairs of similar sentences according to a predetermined similarity criterion for extracting synonym expressions in the synonym expression extraction step, and the computer extracts the extracted synonym expressions. A similarity criterion correction step for correcting the similarity criterion for extracting the synonym expression based on the correctness / incorrect information indicating the correctness and the similarity of the synonym expression may be included.

Synonymous expression extraction program according to the present invention is a synonymous expression extraction program for extracting a synonymous expression from similar sentence pairs paired sentences similar to each other, the computer included in each sentence similar sentence pairs Structure relation between common expressions, which are expressed expressions, and a structure relation specifying process for specifying a structure relation between a difference expression and a common expression, which is an expression included only in each sentence of a similar sentence pair, and the specified structure relation Based on the above, a similarity calculation process for obtaining a similarity of a synonym expression candidate using a difference expression included in a similar sentence pair as a candidate for a synonym expression, and a common expression group included in each sentence of the similar sentence pair in common The relative positional relationship that is similar in the sentence structure of each sentence, and the difference expression group and the common expression group that are different between sentences of similar sentence pairs are similar in the sentence structure of each sentence Based on that the positional relationship extraction, to execute the same meaning expression extracting process of extracting the synonymous expression from similar sentence pair, synonymous expression extraction processing, based on the calculated similarity, the candidate of the synonymous expression as synonymous expression This is to execute a process for determining whether or not to perform the process .
Another aspect of the program for extracting synonymous expressions according to the present invention is a program for extracting synonymous expressions from a pair of similar sentences in which sentences similar to each other are paired. Based on the sentence structure identifying process that identifies the sentence structure, the common expression that is commonly included in each sentence of the similar sentence pair, and the expression that is included only in each sentence of the similar sentence pair, based on the identified sentence structure An expression extraction process for extracting a difference expression, a relative position relationship between the common expressions in the sentence structure of each sentence of the similar sentence pair, and the common expression and the difference expression based on the extracted common expression and the difference expression Based on the positional relationship identification process that identifies the relative positional relationship between and the identified relative positional relationship, the similarity level of the difference expression in the similar sentence pair is calculated from the identity of the structural relationship The degree calculation process and the common expression group included in common in each sentence of the similar sentence pair are in a similar relative positional relationship in the sentence structure of each sentence, and the difference expression that is different between the sentences of the similar sentence pair A synonym expression extraction process for extracting synonym expressions from pairs of similar sentences based on the fact that the group and the common expression group have similar relative positional relationships in the sentence structure of each sentence. This is for executing a process of extracting synonymous expressions based on the calculated similarity.
Still another aspect of the program for extracting synonymous expressions according to the present invention is a program for extracting synonymous expressions for extracting synonymous expressions from a pair of similar sentences in which sentences similar to each other are paired. Included only in each sentence of common sentence and similar sentence pair, which is a common expression included in each sentence of similar sentence pairs, based on the sentence structure identifying process that identifies the sentence structure of Based on the expression extraction process that extracts the difference expression that is the expression, and the extracted common expression and difference expression, the relative positional relationship between the common expression and the difference expression in the sentence structure of each sentence of the similar sentence pair is specified. Based on the identified relative positional relationship, similarity calculation processing for calculating the similarity of the difference expression in the similar text pair from the identity of the structural relationship, and similar text pair A common expression group included in each sentence is in a similar relative positional relationship in the sentence structure of each sentence, and a difference expression group and a common expression group that differ between sentences of similar sentence pairs are each sentence. The synonymous expression extraction process for extracting the synonym expression from the similar sentence pair is executed based on the similar relative positional relationship in the sentence structure, and based on the similarity calculated in the synonym expression extraction process, This is for executing processing for extracting synonymous expressions.

  In addition, the synonym expression extraction program executes a sentence conversion process for converting sentences included in a pair of similar sentences using a predetermined sentence conversion rule so that the sentence structure of similar sentences has a similar structure. It may be allowed.

  The synonym expression extraction program causes the computer to execute a process of extracting a synonym expression from a pair of similar sentences according to a predetermined similarity criterion for extracting the synonym expression in the synonym expression extraction process. A similarity criterion correction process for correcting a similarity criterion for extracting a synonym expression may be executed based on correct / incorrect information indicating whether the extraction result is correct and the similarity of the synonym expression.

  The first synonymous expression extraction system according to the present invention performs dependency analysis on each sentence of a pair of similar sentences, and generates dependency tables indicating dependency relationships between clauses in each sentence. An expression comparison means for extracting a common expression that is a character string that is commonly included in a pair of similar sentences and a difference expression that is a character string that is included only in each sentence and that is a candidate for a synonymous expression, and dependency analysis A structure that indicates which branch, leaf, or root in the tree structure the position information of the common expression and the difference expression in the tree structure of the sentence obtained by A relationship identification table is generated, and using the information of the generated structure relationship identification table, the relative positional relationship between the common representations that are common representation structures in the tree structure, and the common representation and the difference representation that are different representation structures. The structural relationship specifying means for classifying the relative positional relationship of the above into seven types according to a predetermined rule, that is, the same column, the series, the parallel, the front row, the rear row, the row, or other relationships which are the structural relationships, and the common expression structure in the similar sentence pair And the identity of the different representation structures are determined according to an identity determination table that defines the identity of each structural relationship. Based on the identity of each structural relationship, the similarity of the common representation structure in the similar sentence pair and By calculating the similarity of the difference expression structure, and calculating the weighted average of the two similarities for the similarity of the synonym expression candidate pair in the difference expression that is a candidate for the synonym expression in the similar sentence pair Similarity determination means for extracting synonymous expressions by calculating and executing threshold processing is provided.

  According to the above configuration, the similarity of the structural relationship can be calculated from the identity of the structural relationship between the common expressions in the similar text pair, and the configuration of the character string of the similar text pair and the similarity of the grammar can be determined. Therefore, synonymous expressions can be extracted without providing special grammar rules in advance. Further, according to the configuration as described above, it is possible to extract a synonym expression from a pair of similar sentences based only on information about a dictionary required for dependency analysis without requiring a special dictionary or symbol. it can. Furthermore, according to the configuration as described above, the synonym expression can be extracted with high accuracy even if only a pair of similar sentence pairs is used without using statistical processing in accordance with the principle of synonym expression extraction. .

  The second synonym expression extraction system according to the present invention includes, in addition to the components in the first synonym expression extraction system, a tree structure in which similar sentence pairs are similar by the dependency analysis before the dependency analysis is performed. It is characterized by having a text shaping means for shaping the text.

  According to the above configuration, synonymous expressions can be extracted with higher accuracy without losing the advantages of the synonymous expression extracting system that can be realized by the configuration of the first synonymous expression extracting system.

  The third synonymous expression extraction system according to the present invention is a common expression based on which the system user determines the correctness of the extraction result for the extraction result of the synonymous expression extraction system and the similarity of the extraction result. A structure between common expressions that classifies the extraction results into a true synonym expression and a non-true synonym expression based on the similarity of the structural relation between the common expression and the similarity of the structural relation between the common expression and the different expression It is characterized by comprising similarity reference correction means for estimating an appropriate weighting factor for the similarity of the relationship and the similarity of the structural relationship between the common expression and the difference expression.

  According to the above configuration, the similarity criterion parameter for extracting the synonym expression is corrected to an appropriate value without losing the advantage of the synonym expression extraction system that can be realized by the configuration of the first synonym expression extraction system. And the accuracy of synonymous expression extraction can be improved.

  According to the present invention, the common expression group that is commonly included in each sentence of the similar sentence pair has a similar relative positional relationship in the sentence structure of each sentence, and the difference that differs between the sentences of the similar sentence pair Based on the fact that the expression group and the common expression group have similar relative positional relationships in the sentence structure of each sentence, the grammatical similarity of the similar sentence pair can be estimated, and the synonymous expression can be extracted from the similar sentence pair. That is, according to the present invention, it is not determined whether or not the synonym expression is based on only the similarity of the different expressions, but based on the structural relationship between both the common expression and the different expressions included in the similar sentence pair. Thus, it can be determined whether the expression is synonymous. Therefore, synonymous expressions can be extracted from similar sentence pairs without providing special grammar rules in advance.

  In addition, according to the present invention, by simplifying the relative positional relationship of each expression using a dependency table indicating the structural relationship between the common expression and the different expression, the number of modifiers in similar sentence pairs can be increased or decreased, the marker word Synonymous expressions can be extracted without being affected by differences in word order. Furthermore, since it is determined whether or not it is a synonymous expression using a criterion that can design the identity of the relative positional relationship in similar sentence pairs, the strength of the grammatical constraints according to the sentence pair from which synonymous expressions are extracted It is also possible to change.

  Moreover, according to this invention, a synonym expression can be extracted only based on the relationship of the character string in a similar sentence pair. Further, according to the present invention, although a dictionary is used in morphological analysis and dependency analysis, it is sufficient to use only a dictionary that is generally widely used, and synonymous expressions can be extracted without preparing a special dictionary in advance. Can do. Therefore, synonymous expressions can be efficiently extracted from pairs of similar sentences without using special dictionary preparation or symbol expressions.

  Further, according to the present invention, using a method based on the principle of synonymous expression extraction, it is highly accurate whether or not a candidate for synonymous expressions is truly a synonymous expression even if there are only a few similar sentence pairs. The synonymous expression can be extracted with high accuracy. In addition, synonymous expressions can be extracted without using clustering or processing that requires statistical conclusions. Therefore, even if there are only a few similar sentences, synonymous expressions can be extracted with high accuracy.

  The best mode for carrying out the present invention will be described below with reference to the drawings.

Embodiment 1 FIG.
First, a first embodiment of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram showing an example of the configuration of a synonymous expression extraction system according to the present invention. As shown in FIG. 1, the synonymous expression extraction system 10 includes dependency analysis means 101, expression comparison means 102, structural relationship identification means 103, and similarity calculation means 104. The synonym expression extraction system 10 includes a similar sentence pair database 501 and an output unit 502.

  FIG. 2 is a flowchart illustrating an example of processing in which the synonym expression extraction system 10 extracts synonym expressions. Hereinafter, the function of each component of the synonym expression extraction system 10 and the operation of the synonym expression extraction system 10 will be described with reference to the block diagram shown in FIG. 1 and the flowchart shown in FIG.

  In the present embodiment, the synonymous expression extraction system 10 is specifically realized by a computer (for example, an information processing apparatus such as a personal computer) that operates according to a program. The synonym expression extraction system 10 is applied to, for example, applications such as an information search system that executes a synonym expression search, a project management system that performs project management, and a document management system that manages electronic documents.

  In the example illustrated in FIG. 1, the similar sentence pair database 501 and the output unit 502 are provided outside the synonym expression extraction system 10, but the similar sentence pair database 501 and the synonym expression extraction system 10 are provided. Output means 502 may be provided.

  Specifically, the similar text pair database 501 is realized by a database device such as a magnetic disk device or an optical disk device. The similar sentence pair database 501 stores in advance two or more sentences in a single language whose meanings are similar to each other. For example, as similar sentence pairs, sentences may be extracted from different news articles described with respect to the same content and stored in the similar sentence pair database 501. In addition, for example, the similar sentence pair database 501 may store a plurality of sentences translated from sentences written in the same language (foreign language) as similar sentence pairs, or from electronic documents having a reference relationship with each other. You may memorize | store the extracted sentence as a similar sentence pair.

  In the present embodiment, similar sentence pairs extracted by a similar sentence pair extraction system (not shown) for extracting similar sentence pairs from an electronic document or the like are stored in advance in the similar sentence pair database 501. For example, the similar sentence pair extraction system automatically extracts similar sentence pairs from Web documents on the Internet, electronic documents stored in a common server, etc., and stores them in the similar sentence pair database 501 at predetermined time intervals. Further, for example, the similar sentence pair database 501 may store in advance similar sentence pairs created by the user.

  FIG. 3 is an explanatory diagram showing an example of similar sentence pairs stored in the similar sentence pair database 501. In the similar sentence pair shown in FIG. 3, it is assumed that the expression “ABC function” and the expression “XYZ function” are synonymous expressions. Hereinafter, a method for extracting synonymous expressions will be described using the similar sentence pair shown in FIG. 3 as an example. In the present embodiment, a case where a synonymous expression between two similar sentences is extracted will be described as an example. When three or more similar sentences are included in a pair of similar sentences, a pair of similar sentences obtained by combining two similar sentences may be created, and the process may be executed for all the combinations.

  The synonym expression extraction system 10 starts synonym expression extraction processing in accordance with an operation by the user. For example, the synonym expression extraction system 10 displays a display screen including a form for synonym expression extraction on the output unit 502 that is a display device or the like. In this case, when the execution button is clicked by the user from the synonym expression extraction form, the synonym expression extraction system 10 starts the synonym expression extraction process.

  Specifically, the dependency analysis unit 101 is realized by a CPU of an information processing apparatus that operates according to a program. When the synonym expression extraction process is started, the dependency analysis unit 101 first extracts a similar sentence pair from the similar sentence pair database 501. Next, the dependency analysis unit 101 performs dependency analysis on each sentence included in the extracted similar sentence pair, and outputs the dependency analysis result to the expression comparison unit 102 (step S704a).

  In general, before performing dependency analysis, it is necessary to perform a morpheme analysis process that decomposes a sentence into morphemes. In the present embodiment, it is assumed that the morphological analysis process is incorporated in the dependency analysis unit 101, and the dependency analysis unit 101 executes the dependency analysis after executing the morphological analysis process in step S704a. In addition, you may make it provide the means which performs a morphological analysis process separately from the dependency analysis means 101 which performs a dependency analysis process.

  The dependency analysis unit 101 performs morphological analysis using an existing technique such as HMM (Hidden Markov Model). Further, the dependency analysis unit 101 executes dependency analysis using existing techniques such as a probability model and syntax analysis.

  FIG. 4 is an explanatory diagram illustrating the dependency relationship between each phrase by performing dependency analysis on each sentence shown in FIG. 3. In FIG. 4, the node represents a phrase, and the tip of the arrow represents the phrase destination. As shown in FIG. 4, the dependency analysis unit 101 can express each sentence using a tree structure having clauses as nodes by executing dependency analysis on each sentence of the similar sentence pair. . Here, the dependency tree structure shown in FIG. 4 is called a dependency structure.

  The dependency analysis unit 101 stores the morpheme analysis result and the information on the correspondence between each phrase in a recording medium in a data format as shown in FIG. 5, for example. For example, the dependency analysis unit 101 stores, as a storage medium, information on the correspondence between morphological analysis results and phrases in a memory of an information processing device, a hard disk device, a CD-ROM, a DVD-ROM, a flexible disk, or the like. As shown in FIG. 4, the data is stored in a table format.

  As shown in FIG. 5, the table showing the morphological analysis results and the correspondence information of each clause includes the word extracted by the morphological analysis, the part of speech of the word, and the phrase ID obtained by the dependency analysis. It is. The phrase ID is identification information for specifying a phrase. The same phrase ID is assigned to words belonging to the same phrase.

  Further, the dependency analysis unit 101 stores the obtained dependency structure in a storage medium in a table format in a data format as shown in FIG. 6, for example. Hereinafter, the table showing the dependency structure shown in FIG. 6 is referred to as a dependency table. In the dependency table shown in FIG. 6, the “phrase ID” in the left column corresponds to the phrase ID shown in FIG. The “phrase” in the center column is a character string that forms one clause. “Destination ID” in the right column indicates the phrase ID of the related phrase related to the phrase.

  For example, in the example illustrated in FIG. 6, it is understood that the phrase “moving image processing” with the phrase ID 001 relates to the phrase “for” with the phrase ID 002. Further, when there is no such clause, for example, the relation ID is 000 so that it can be determined that there is no relation.

  Specifically, the expression comparison unit 102 is realized by a CPU of an information processing apparatus that operates according to a program. The expression comparison unit 102 inputs the analysis result from the dependency analysis unit 101. Further, the expression comparison unit 102 identifies the common expression and the difference expression in the similar sentence pair based on the input dependency analysis result, and outputs the identification result of the common expression and the difference expression to the structural relationship extraction unit 103. (Step S704b).

  The expression comparison unit 102 may arbitrarily select the part of speech of the target expression to be extracted as a common expression or a different expression. For example, the expression comparison unit 102 may extract a common expression or a difference expression for a phrase including a noun. In addition, the expression comparison unit 102 may extract a common expression or a difference expression for a phrase including an independent word. Furthermore, the expression comparison unit 102 not only extracts by phrase breaks, but also extracts common expressions and difference expressions by using a phrase including a noun and one or more phrase groups that modify the phrase as one expression. Good. In addition, the expression comparison unit 102 may extract a common expression or a difference expression in units of subtrees with dependency structures. Note that the expression comparison unit 102 can extract the part of speech of each word included in each phrase based on the morphological analysis result shown in FIG. 5 and information on the correspondence between each phrase.

  Hereinafter, in order to make the explanation easy to understand, a process of extracting synonymous expressions in the case where expressions extracted as common expressions or different expressions are used as noun phrases included in each phrase will be described. In general, when performing an information search or the like, an information search operation is often performed by inputting a noun or a noun phrase. Also, in general, there are many synonymous expressions of verbs, adjectives, etc. that can be recognized immediately when the user sees them, but if nouns or noun phrases contain synonyms, In many cases, it is impossible to immediately recognize whether a user is synonymous or not just by looking at a noun phrase. Therefore, when synonyms are included in nouns and noun phrases between electronic documents, etc., there is a high possibility of being an obstacle to information retrieval. Therefore, in the present embodiment, a case will be described in which synonymous expressions in noun phrases are extracted from similar sentence pairs. The synonymous expression extraction system 10 is not limited to nouns and noun phrases, and may extract synonymous expressions such as verbs and adjectives.

  Hereinafter, each phrase is also referred to as an expression. In the example illustrated in FIG. 4, the expression comparison unit 102 extracts “high speed drawing” (706a, 706d) and “moving image processing” (706b, 706e) as common expressions. In FIG. 4, the square frame indicated by shading indicates that a difference expression is included. That is, the expression comparison unit 102 extracts “ABC function” 706c, “XYZ function” 706f, and “visualization diagnosis” 706g as difference expressions.

  Specifically, the structural relationship specifying unit 103 is realized by a CPU of an information processing apparatus that operates according to a program. The structural relation specifying means 103 specifies the structural relation between the common expressions in each sentence and the structural relation between the common expressions and the different expressions based on the common expressions and the different expressions extracted from the pair of similar sentences. Each identified structural relationship is output to the similarity determination means 104 (step S704c). The structural relationship is defined by the relative positional relationship of each expression in the dependency structure. In the dependency structure, the structural relationship can be divided into two types according to whether a reference expression (hereinafter referred to as a reference expression) exists at a non-branching position or a branching position. An expression that specifies a structural relationship with respect to a reference expression is called a target expression. An expression located at a branch is called a branch expression.

  FIG. 7 is an explanatory diagram illustrating an example of a structural relationship between common expressions and a structural relationship between common expressions and different expressions. FIG. 7A shows the structural relationship when the reference expression (709a) is located in a non-branch. FIG. 7B shows the structural relationship when the reference expression (709b) is located at a branch. Further, in FIG. 7, a circle represents one expression, and an arrow represents a dependency. The ● mark represents the standard expression.

  When the reference expression is located at a non-branch, three structural relationships of the same column, series, and parallel are defined by the position of the target expression indicated by the broken-line square frame shown in FIG. In this case, the structural relationship specifying unit 103 specifies whether the target expression is the same, serial, or parallel. Here, the same column is a set of expressions that have a direct dependency relationship with the reference expression, and the series is an expression in which the reference expression is “dependency” or “reception” when viewed from the overall dependency structure of the sentence. It is a set, and parallel is a set of expressions related to a common branch expression.

  On the other hand, when the reference expression is located at a branch, the three structural relationships of the front row, the rear row, and the horizontal row are defined according to the position of the target representation indicated by the broken-line square frame shown in FIG. In this case, the structural relationship specifying unit 103 specifies whether the target expression is a front row, a back row, or a row. Here, the front row is a set of expressions that are “involved” with respect to the reference expression, the back row is a set of expressions that are “received” with respect to the reference expression, and the row is related to a common branch expression. A collection of expressions.

  In the dependency structure, if there are two or more branch expressions between the expressions specifying the structural relationship, the structural relationship specifying unit 103 determines that the structure does not belong to any of the six relationships shown above, and the structure Identify the relationship as “Other”.

  Next, taking the sentence 2 as an example, the process in which the structural relationship identifying unit 103 identifies the structural relationship will be described. In the structure relation specifying process, the structure relation specifying unit 103 first generates the structure relation specifying table shown in FIG. 8 based on the dependency table shown in FIG. The structural relationship specification table is information that summarizes information necessary for specifying the structural relationship. As shown in FIG. 8, the structure relationship specification table includes a phrase ID, a position (branch / non-branch) in the dependency structure of the expression corresponding to the phrase ID, and the proximity of each expression on the dependency side and the reception side. It includes a phrase ID for specifying the branch expression (an engagement side adjacent branch ID, a receiving side adjacent branch ID).

  As shown in FIG. 9, the dependency side means the leaf side in the dependency structure, and the reception side means the root side in the dependency structure. The dependency side adjacent branch ID and the reception side adjacent branch ID are identification information for grasping which branch expression each expression is sandwiched in the dependency structure. However, in the dependency structure, the expression corresponding to the root does not have the reception side adjacent branch ID, and the expression in the same row as the leaf does not have the dependency side adjacent branch ID. Therefore, as shown in FIG. 9, a leaf ID and a root ID are used instead of the phrase ID of the branch expression. By doing so, the structural relationship specifying unit 103 assigns any of the phrase ID, the leaf ID, and the root ID of the branch expression to the relationship-side adjacent branch ID and the receiver-side adjacent branch ID of any expression.

  The structural relationship identification unit 103 generates a structural relationship identification table according to the following processing. Based on the dependency table shown in FIG. 6, the structure relation specifying unit 103 determines that the expression is a branch expression if there are two or more same phrase IDs in the dependency ID, and is not branched if the same phrase ID is one or less. The expression of Through such processing, the structural relationship specifying unit 103 specifies branch / non-branch of each expression.

  Next, the structural relationship specifying unit 103 extracts a phrase ID that does not exist in the dependency ID in the dependency table. The extracted phrase ID expression corresponds to a leaf in the dependency structure, and the structure relation specifying unit 103 assigns a leaf ID to the dependency side adjacent branch ID of the extracted phrase ID expression. In this case, the structural relationship specifying unit 103 uses a different value as each leaf ID.

  Next, the structure relation specifying unit 103 sequentially refers to each expression from the expression corresponding to the leaf until the relation expression becomes a branch expression, and the expression appearing in the tracing process is stored in a memory or the like. Store in a storage medium. Further, the structural relationship specifying unit 103 assigns a relationship-side proximity branch ID of an expression (corresponding to a leaf) whose reference is started to the relationship-side proximity branch ID of the expression stored in the storage medium. Further, the structure relation specifying unit 103 assigns the phrase ID of the branch expression described above as the receiving side adjacent branch ID. Further, the structure relation specifying unit 103 assigns the relation side adjacent branch ID of the expression directly related to the expression as the relation side adjacent branch ID of the branch expression.

  The structural relationship specifying unit 103 assigns two or more IDs as the dependency side adjacent branch IDs of the branch expression. If the branch expression is not found and becomes the root of the dependency dependency structure, the structure relation specifying unit 103 sets the receiving branch expression ID of the expression as the root ID and ends the process. If this is not the case, the structural relationship specifying means 103 sequentially refers to the branch expression from the branch expression until the branch expression becomes the branch expression, and the expression that appears in the trace process is stored in a storage medium such as a memory. Remember me. Further, the structure relation specifying unit 103 gives the phrase ID of the branch expression that has been referred to the relation-side adjacent branch ID of the expression stored in the storage medium.

  When the above processing is repeatedly executed and the reference destination becomes the root of the dependency structure, the structural relationship specifying unit 103 assigns the root ID to the receiving side branch expression ID of the expression and ends the processing. FIG. 8 is an explanatory diagram showing an example of the structural relationship specification table obtained based on the dependency table shown in FIG.

  Next, the structural relationship specifying unit 103 specifies the structural relationship for two expressions including a common expression or a different expression in the structure relationship specifying table. When the reference expression is non-branching, the structural relationship specifying means 103 (1) If the receiving side adjacent branch ID and the related side adjacent branch ID are the same in the two expressions, the structural relationship is set to “same row”. Is identified. In addition, the structural relationship specifying unit 103 specifies (2) the structural relationship as “parallel” if the receiving side adjacent branch IDs are the same and the related side adjacent branch IDs are different in the two expressions. Further, the structural relationship specifying means 103 specifies (3) the structural relationship as “series” when one receiving side adjacent branch ID is the same as the other related side adjacent branch ID. Further, the structural relationship specifying unit 103 specifies (4) the structural relationship as “others” if there is no match between the receiving side adjacent branch ID and the relationship side adjacent branch ID of the two expressions.

  Further, when the reference expression is a branch expression, the structure relation specifying unit 103 specifies (1) the structure relation as “front row” when the receiving side adjacent branch ID of the target expression is the phrase ID of the reference expression. . In addition, the structural relationship specifying unit 103 specifies (2) the structural relationship as “back row” when the relationship-side adjacent branch ID of the target expression is the phrase ID of the reference expression. Further, the structural relationship specifying means 103 specifies (3) the structural relationship as “parallel” if the receiving side adjacent branch IDs of the two expressions are the same. Further, the structural relationship specifying unit 103 specifies (4) the structural relationship as “others” if there is no match between the receiving side adjacent branch ID and the relationship side adjacent branch ID of the two expressions.

  Note that the above-described method of specifying the structural relationship is an example, and the structural relationship specifying unit 103 may specify the structural relationship using another method as long as the desired structural relationship can be specified. Hereinafter, the structural relationship between the common representations is referred to as a common representation structure, and the structural relationship between the common representation and the difference representation is referred to as a difference representation structure.

  In the two expressions for obtaining the structural relationship, when one is a branching expression and the other is not a branching expression, the structural relation changes depending on which is used as a reference expression. Therefore, in the present embodiment, the structural relationship specifying unit 103 specifies the two structural relationships by executing processing by exchanging the reference expression and the target expression between the two expressions. In this way, the accuracy of synonymous expression extraction can be further improved by specifying and processing two types of structural relationships in which the reference expression and the target expression are interchanged. Note that the synonymous expression extraction process may be performed using only one of the expressions as the reference expression and the other as the object expression without replacing the reference expression and the object expression.

  When there are m common expressions in a pair of similar sentences, the structure relation specifying unit 103 specifies m × (m−1) common expression structures in each sentence. Further, the structural relationship specifying unit 103 specifies m × 2 difference expression structures for one difference expression. For example, when comparing sentence 1 and sentence 2, since m = 2, the structural relationship specifying unit 103 specifies two common expression structures. Also, for sentence 1 and sentence 2, there are one and two difference expressions, respectively, so the structural relationship specifying means 103 specifies four difference expression structures in sentence 1, and eight differences in sentence 2 Specify the representation structure.

  FIG. 10 is an explanatory diagram showing the result of the structural relationship specifying unit 103 specifying the structural relationship for the sentence 1. The identification result shown on the right side of FIG. 10 is a structural relationship obtained by inverting (replacement) the reference expression and the target expression used in the identification result shown on the left side of FIG. In FIG. 10, the origin of the broken line arrow indicates the reference expression, and the tip of the wavy arrow indicates the target expression. In FIG. 10, expressions (sentences) including the common expression are indicated by white characters, and expressions including the difference expression are illustrated as a shaded square frame.

  As shown in FIG. 10, based on the structural relationship identification result by the structural relationship identifying unit 103, the sentence 1 has a common expression structure between “high-speed rendering” and “moving image processing” regardless of the difference in the standard expression. It turns out that it is "parallel" (720a, 820a). It can also be seen that the difference representation structure between the “ABC function” and the “moving image processing” is “parallel” (720b, 820b). It can also be seen that the difference representation structure between the “ABC function” and the “high-speed drawing” is “same column” (720c, 820c).

  FIG. 11 is an explanatory diagram showing the result of the structural relationship specifying unit 103 specifying the structural relationship for the sentence 2. Note that the identification result shown on the lower side of FIG. 11 is a structural relationship obtained by inverting (replacement) the reference expression and the target expression used in the identification result shown on the upper side of FIG. 11 is the same as that shown in FIG.

  As shown in FIG. 11, based on the structural relationship specifying result by the structural relationship specifying means 103, the common expression structure of “high-speed rendering” and “moving image processing” is also used in the sentence 2 regardless of the difference in the standard expression. It turns out that it is "parallel" (730a, 830a). Also, the difference representation structure between the “XYZ function” and “moving image processing” is “parallel” (713b, 830b), and the difference representation structure between the “XYZ function” and “high-speed rendering” is “same row” (730c, 830c). I know that there is. Also, the difference representation structure between “visualization diagnosis” and “high-speed rendering” and the difference representation structure between “visualization diagnosis” and “moving image processing” include two or more branch expressions (“used” and “ It is understood that the structural relationship is “others” (730d, 730e, 830d, 830e).

  Specifically, the similarity determination unit 104 is realized by a CPU of an information processing apparatus that operates according to a program. The similarity determination means 104 is different between the sentences of the similar sentence pair, and the common expression group included in common in each sentence of the similar sentence pair has a similar relative positional relationship in the sentence structure of each sentence. A function is provided for extracting synonymous expressions from pairs of similar sentences based on the fact that the difference expression group and the common expression group are in a similar relative positional relationship in the sentence structure of each sentence.

  The similarity determination unit 104 receives a similar sentence pair from the structural relationship extraction unit 103 and determines the identity of the common expression structure and the different expression structure in the input similar sentence pair (step S704d). Further, the similarity determination unit 104 extracts the synonym expression by calculating the similarity of the synonym expression candidate pair in the similar sentence pair based on the determination result of the identity, and outputs the extraction result of the synonym expression to the output unit 502. Output (step S704e).

  The synonymous expression candidate pair is information obtained by pairing different expressions that are candidates for synonymous expressions in a pair of similar sentences. When ni and nj different expressions are included in the sentence i and the sentence j, which are similar sentence pairs, there are ni × nj combinations of synonymous expression candidate pairs. The similarity determination unit 104 calculates the similarity for all of these combinations of synonym expression candidate pairs.

  First, the similarity determination unit 104 identifies a comparative structure relationship in the synonymous expression candidate pair. The comparative structure relationship is a common expression structure and a different expression structure to be compared in order to calculate the similarity between synonymous expression candidate pairs. The similarity determination means 104 compares the common expression structures with the same reference expression and target expression in the similar sentence pairs. In addition, the similarity determination unit 104 compares different expression structures that have the same common expression as the reference expression in the similar sentence pair or the same common expression as the target expression.

  FIG. 12 shows an example of the comparative structure relationship when the number (m) of common expressions in similar sentence pairs is two. In FIG. 12, common expressions 1 and 2 are expressions of the same character string used in sentences i and j, and difference expression a and difference expression b are synonymous expression candidate pairs. Further, the synonym expression candidate pairs are indicated by shading in FIG. In the example shown in FIG. 12, the number of structural relations compared by the similarity determination unit 104 is 2 for the common expression structure and 4 for the different expression structure for one synonym expression candidate pair, for a total of 6 It becomes a piece. That is, the similarity determination unit 104 calculates the similarity of the synonymous expression candidate pairs based on the identity of the six structural relationships.

  FIG. 13 shows a comparative structure relationship for the synonymous expression candidate pair in the sentence 1 and the sentence 2. In sentence 1 and sentence 2, there are two synonymous expression candidate pairs: “ABC function” and “XYZ function”, and “ABC function” and “visualization diagnosis”. Similarity determination means 104 compares two common expression structures and four different expression structures for both of the synonymous expression candidate pairs, and determines the identity of the structural relationship, as in FIG. Then, the similarity of the synonymous expression candidate pair is calculated.

  In the similar sentence pair, when it is guaranteed that the grammatical structure in each sentence and the word order of the common expression and the difference expression are equivalent, the similarity determination unit 104 only determines the identity of the difference expression structure. May be used to calculate the similarity of the synonymous expression candidate pair. In this case, in the example illustrated in FIG. 13, the similarity determination unit 104 compares the four different representation structures with each of the synonymous expression candidate pairs of “ABC function” and “XYZ function” and “ABC function” and “visualization diagnosis”. Then, the identity of the structural relationship is determined, and the similarity of the synonymous expression candidate pair is calculated.

  Next, a determination method of the determination process of the identity of each structural relationship executed by the similarity determination unit 104 will be described. FIG. 14 is an explanatory diagram illustrating an example of a table indicating a determination criterion regarding the identity of the structural relationship. The table shown in FIG. 14 sets whether all combinations of two structural relationships are regarded as the same. Hereinafter, the table shown in FIG. 14 is referred to as an identity determination table. Note that the identity determination table illustrated in FIG. 14 is stored in advance in a storage medium such as a hard disk device or a memory included in the information processing apparatus, for example. Further, since the identity determination table shown in FIG. 14 is a target matrix, it is assumed that no value is included in the lower triangular matrix portion.

  In the identity determination table shown in FIG. 14, determination values for determining three types of identity of “◯”, “×”, or “order” are set for each combination of structural relationships. Here, “◯” means that the comparison structural relationships are determined to be the same, and “x” means that they are determined to be different. Further, “order” represents that, in the comparative structure relationship, when the order in the sentences of the reference expression and the target expression is the same, they are determined to be the same.

  The similarity determination unit 104 determines the identity of each structural relationship according to the identity determination table shown in FIG. In this case, the similarity determination unit 104, as a general rule, when the comparison structural relationship is the same (corresponding to the diagonal elements shown in FIG. 14), except for the case where the structural relationship is “other” It can be determined that the structural relationship is the same. However, in the identity determination using the identity determination table, it is allowed to be determined to be the same even when the structural relationship does not match.

  For example, in the identity determination table, if the comparison structure relationship is “parallel” and “row” in the similar sentence pair, it is assumed that they are the same “◯” 760a. Although the row and the parallel are different in the position of the reference expression, they both define “relationship between expressions related to the same branch expression”, and can be determined to have substantially the same structure.

  Further, by setting “order” as the determination value in the identity determination table, the following processing can be realized. FIG. 15 is an explanatory diagram showing a sentence 3 and a sentence 4 which are other examples of similar sentence pairs. In the similar sentence pair shown in FIG. 15, it is assumed that “DEF function” and “KLM function” are synonymous expressions. In the example illustrated in FIG. 15, the expression comparison unit 102 identifies “product” and “new development” as common expressions, “DEF function” of sentence 3, “KLM function”, and “moving image processing” of sentence 4. Is specified as a difference expression. Therefore, in the example illustrated in FIG. 15, “DEF function” and “KLM function”, “DEF function”, and “moving image processing” are synonymous expression candidate pairs.

  FIG. 16 is an explanatory diagram showing a dependency structure of similar sentence pairs shown in FIG. In each structural relationship shown in FIG. 16, the source of the arrow indicates the reference expression, and the tip of the arrow indicates the target expression. Also, the structural relationship “same row” 780a and “series” (780b, 780c) shown in FIG. 16 are comparative structural relationships. In this case, “same row” 780a in sentence 3 and “series” 780b in sentence 4 do not match. Therefore, the conventional synonym expression extraction method cannot extract synonym expressions from the sentence 3 and the sentence 4.

  However, in the present embodiment, even in the case shown in FIG. 16, the similarity determination unit 104 determines that the “same line” 780a in the sentence 3 and the “series” in the sentence 4 are based on the identity determination table shown in FIG. It can be determined that the difference representation structural relationship with “780b” is the same. That is, in the identity determination table shown in FIG. 14, the sameness determination criterion of “same column” and “series” is set to “order”, and if the common expression and the target expression match in order, the difference expression It can be determined that the structural relationship is the same. In the example illustrated in FIG. 16, in the sentence 3, the reference expression “DEF function” precedes the target expression “product” in the word order, and also in the sentence 4, the reference expression “KLM function” is higher than the target expression “product”. It comes before the word order, and the order of the common expression and the target expression is the same. Therefore, the similarity determination unit 104 can determine that the difference representation structure relationship between the “same column” 780a and the “series” 780b is the same.

  On the other hand, in FIG. 16, as with the “KML function”, “moving image processing” has a relationship of “serial” 780c with the target expression “product”, but the similarity determination unit 104 has a structural relationship of “serial” 780c. The structural relationship “same” 780a between the “DEF function” and “product” can be determined to be different. That is, in the example shown in FIG. 16, the reference expression “moving image processing” is behind the target expression “product” in the sentence 4 as the word order, and the word order of the reference expression “DEF function” and the target expression “product” in the sentence 3 Therefore, the similarity determination unit 104 determines that the difference representation structure relationship between the “series” 780c and the “same column” 780a is different.

  By executing the above processing, it is possible to make a difference in the structural relationship between the “moving image processing” and the “KML function” with respect to the “DEF function”.

  The identity determination table shown in FIG. 14 is an example, and each element in the identity determination table may be changed by the system user. For example, in the identity determination table shown in FIG. 14, the restriction on the word order may be relaxed by changing “order” to “◯”.

  FIG. 17 and FIG. 18 show the identity of the comparison structure relationship between the synonymous expression candidate pairs “ABC function” and “XYZ function”, “ABC function”, and “visualization diagnosis” specified by the similarity determination unit 104, respectively. It is explanatory drawing shown. In FIG. 17 and FIG. 18, the comparative structural relationship is indicated by a bidirectional arrow, and “same” and “difference” given to the arrows indicate the determination results based on the identity determination table of FIG. In the examples shown in FIGS. 17 and 18, it can be seen that the “ABC function” and the “XYZ function” have the same comparative structural relationship. In addition, it can be seen that “ABC function” and “visualization diagnosis” are different in four different representation structures.

  Next, the similarity determination unit 104 calculates the similarity of the synonymous expression candidate pair based on the obtained identity of the structural relationship. Note that the similarity determination unit 104 executes the process for obtaining the similarity of the synonym expression candidate pairs for all the synonym expression candidate pairs. In this case, the similarity determination unit 104 may calculate the similarity of the synonym expression candidate pair by using, for example, Expression (1) as a similarity calculation method.

Similarity = (same number of structural relationships) ÷ (number of structural relationships compared) Equation (1)

  The similarity determination unit 104 determines whether or not the similarity calculated using Expression (1) is equal to or greater than a predetermined threshold (step S704f). If the degree of similarity is greater than or equal to a predetermined threshold, the similarity determination unit 104 determines that each synonym expression candidate is similar and determines that it is a synonym expression. And the similarity determination means 104 extracts as a synonymous expression from a similar sentence pair (step S704g). If the similarity is not greater than or equal to the predetermined threshold, if there is a next similar sentence pair, the process proceeds to synonymous expression extraction processing for the next similar sentence pair (step S704h).

  In the example shown in FIG. 17 and FIG. 18, the “ABC function” and the “XYZ function” have six comparison structural relationships (two common representation structures and four different representation structures), and all of these structural relationships. Are similar to each other, the similarity determination unit 104 obtains the similarity as 6/6 = 1.0. In addition, since “ABC function” and “visualization diagnosis” are the same only in the common expression structure, the similarity determination unit 104 obtains the similarity as 2/6 = 0.33. For example, if the threshold value is 0.8, the similarity determination unit 104 determines that the “ABC function” and the “XYZ function” are similar and extracts them as synonymous expressions.

  Further, the similarity determination unit 104 performs weighting on the structural relationship between the common expressions and the identity of the structural relationship between the different expressions and the common expression, and obtains a weighted average to calculate the similarity. May be performed. In this case, for example, the similarity determination unit 104 calculates the degree of similarity by obtaining a weighted average using Equation (2).

Similarity = w × (similarity of common expression structure) + (1−w) × (similarity of different expression structure)
Formula (2)

  However, the similarity determination unit 104 obtains the similarity of the common expression structure and the different expression structure using the expressions (3) and (4), respectively, when the similarity is obtained using the expression (2).

(Similarity of common expression structures) = (same number of common expression structures) ÷ (number of common expression structures compared)
Formula (3)

(Similarity of difference representation structures) = (same number of difference representation structures) ÷ (number of difference representation structures compared)
Formula (4)

  In Expression (2), w is a weighting coefficient indicating how much importance is placed on the identity between common expressions in similar sentence pairs, and takes a value of 0 to 1. Further, if w = (m−1) / (m + 1), Expression (2) is equivalent to Expression (1). Here, m is the number of common expressions. That is, equation (2) can be understood as a general equation including equation (1). If the weighting factor w is 0.2, synonymous expression extraction can be performed with an emphasis on the difference expression structure.

  FIG. 19 is an explanatory diagram showing an example of the degree of similarity using the weighted average obtained by the similarity determination unit 104. FIG. 19 shows a case where the similarity of the common expression structure and the similarity of the difference expression structure in sentence 1 and sentence 2 are calculated using equations (3) and (4), and w = 0.2. The similarity is shown.

  Note that, when there are many synonymous expression candidate pairs that are equal to or greater than the threshold, the similarity determination unit 104 may extract the one having the highest similarity as the synonym expression, or synonymous expressions that are equal to or greater than the threshold. May be output as Further, the similarity determination unit 104 may output only a predetermined number of candidates having higher similarity as synonymous expressions.

  Furthermore, when the structural relationship between the common expressions is different, the similarity determination unit 104 may determine that the structure of the sentence has changed greatly in the similar sentence pair, and may change the threshold value for determining the synonymous expression. In addition, the similarity determination unit 104 detects a portion where the difference between the similarities is statistically large based on the distribution of similarity values when the similarity to a plurality of synonym expression candidate pairs is obtained, and a threshold value May be determined. When there is no similar synonym expression candidate pair, the similarity is calculated again when “others” in the identity determination table shown in FIG. You may recalculate. In this case, synonymous expressions that are similar in the structural relationship of “others” can be extracted.

  It should be noted that the similarity determination means 104 has a fixed value for the similarity of the structural relationship between the common expressions in a pair of similar text pairs. It is not necessary to calculate the degree of similarity between the structural relationships.

  Further, only the extracted synonym expression may be output to the output unit 502 according to the instruction of the similarity determination unit 104, or an expression related to the synonym expression may be output together with the synonym expression. For example, when the similarity determination unit 104 extracts synonymous expressions “ABC function” and “XYZ function” and the expression “XYZ function” is related to the expression “XX company”, the output is output. You may make the means 502 output "ABC function" and "XX company's XYZ function" as synonymous expressions. In this case, the similarity determination unit 104 excludes “XX company's” related to the XYZ function from the synonym expression when the word “XX company” exists in the sentence including the ABC function. May be output to the output unit 502. By doing so, it is possible to extract only expressions related to synonymous expressions that are not included in the sentences to be compared in the similar sentence pair.

  Specifically, the output unit 502 is realized by a display device such as a display device. For example, when the output unit 502 is a display device, the synonym expression extracted by the synonym expression extraction device 10 is displayed in accordance with an instruction from the similarity determination unit 104. Note that the output unit 502 may output the extraction result of the synonym expression as a file, or may notify (send) the system user's terminal using e-mail (e-mail) or the like.

  FIG. 20 is an explanatory diagram illustrating an example of the output result of the synonymous expression output by the output unit 502. As shown in FIG. 20, the output unit 502 outputs what synonym expression is extracted from what sentence according to the instruction of the similarity determination unit 104. In the example shown in FIG. 20, the output unit 502 outputs (displays) that synonymous expressions of ABC function and XYZ function are extracted from sentence 1 and sentence 2.

  The output mode of the output unit 502 may be a display mode other than that illustrated in FIG. For example, as illustrated in FIG. 21, the output unit 502 may display synonym expression candidates and display the similarity calculated together with the synonym expression candidates. As described above, by outputting (displaying) a plurality of candidates for synonymous expressions, it is possible to avoid such an adverse effect that the extracted synonymous expressions are not output due to a slight difference in similarity.

  In the present embodiment, the storage device (not shown) of the information processing apparatus that implements the synonymous expression extraction system 10 stores various programs for extracting synonymous expressions from each sentence included in the similar sentence pair. ing. For example, the storage device of the information processing apparatus has a computer in which a common expression group commonly included in each sentence of a similar sentence pair has a similar relative positional relationship in the sentence structure of each sentence, and the similar sentence pair Execute synonymous expression extraction processing to extract synonymous expressions from pairs of similar sentences based on the fact that the difference expression group and common expression group that differ between the two sentences are in a similar relative positional relationship in the sentence structure of each sentence A synonym expression extraction program for storing the same is stored.

  As described above, according to the present embodiment, the grammatical similarity of a similar sentence pair is estimated based on information such as the similarity of the common expression structure included in the similar sentence pair, and the synonym expression is obtained from the similar sentence pair. Can be extracted. In other words, in the present embodiment, the synonym expression extraction system 10 does not determine whether or not the synonym expression is based on only the similarity of the difference expressions, but the common expression and the difference expression included in the similar sentence pair. Based on both structural relationships, it is determined whether or not the expression is synonymous. Therefore, synonymous expressions can be extracted from similar sentence pairs without providing special grammar rules in advance.

  FIG. 22 is an explanatory diagram showing the principle by which the synonym expression extraction system 10 according to the present invention can extract synonym expressions. In the present embodiment, similar to the conventional synonymous expression extraction method shown in FIG. 32, the same marker word (701a, 901a) exists, and the synonymous expression candidate (701b, 701a, 901a) exists. If the relationship (701c, 901c) with 901b) is similar, the synonym expression candidate is set as a synonym expression. Also, the difference from the conventional synonym expression extraction method shown in FIG. 32 is that, in this embodiment, the synonym expression extraction system 10 is identical in the relationship (701d, 901d) between marker words that are common to similar text pairs. To evaluate. Then, the synonym expression extraction system 10 determines that the grammar of the similar sentence pair, the regularity of the word order, and the like are similar if the common marker words (701a, 901a) are similar.

  By doing as described above, synonymous expressions can be extracted with high accuracy without requiring explicit grammatical constraints in advance for similar sentence pairs. In addition, synonymous expressions can be extracted even when the word order is different between similar sentences, when similar expressions are partially included in the sentences, or when the predicates differ.

  Further, according to the present embodiment, synonymous expressions can be extracted without using a dictionary other than a general dictionary used for dependency analysis. Therefore, similar sentence pairs can be efficiently extracted without preparing a special dictionary or using symbolic expressions.

  Furthermore, according to the present embodiment, as described above, synonym expressions are extracted according to the principle of synonym expression extraction that does not require a large number of similar sentence pairs. Therefore, even from a small number of similar sentence pairs, synonymous expressions can be extracted with high accuracy.

Embodiment 2. FIG.
Next, a second embodiment of the present invention will be described with reference to the drawings. FIG. 23 is a block diagram illustrating a configuration example of the synonymous expression extraction system according to the second embodiment. As shown in FIG. 23, this embodiment is different from the first embodiment in that it includes a text shaping unit 105 in addition to the components shown in FIG.

  Specifically, the text shaping unit 105 is realized by a CPU of an information processing apparatus that operates according to a program. The text shaping unit 105 acquires (extracts) a similar text pair from the similar text pair database 501. In addition, the sentence shaping unit 105 shapes in advance a difference in notation that changes the dependency structure for the extracted similar sentence pair, and outputs the shaped sentence to the dependency analysis unit 101.

  In this embodiment, by shaping the difference in notation in advance, a similar dependency structure can be identified from a pair of similar sentences, and the extraction accuracy of the synonymous expression can be improved. That is, in this embodiment, similar sentences are converted in advance according to a predetermined grammatical rule, so that the dependency structure of sentences can be easily specified, and the synonymous expression extraction accuracy is improved.

  As an example of a grammatical rule that makes the structures similar, the sentence shaping unit 105 converts, for example, “(noun) (noun of sa-changing connection)” to “doing (noun) (noun of sa-changing connection)”. I do. The noun of the S-variable connection is a word that is treated like a verb, accompanied by words such as “to be extracted” or “acquired”, and immediately following “to do” or “to be done”.

  FIG. 24 is an explanatory view showing an example sentence including a noun of a change connection. In the example shown in FIG. 24, it is assumed that the PQR function included in the sentence 5 and the STU function included in the sentence 6 are synonymous expressions. FIG. 25 is an explanatory diagram showing a result of applying dependency analysis to the sentences 5 and 7 shown in FIG. As shown in FIGS. 25A and 25B, the sentence 5 and the sentence 6 are similar sentences, but it can be seen that the structure of the dependency analysis is changed due to the verbalization of the sa variable noun. Therefore, the sentence shaping unit 105 converts the grammar side of the sentence 5 by applying the conversion rule described above, and converts the sentence 5 as shown in the sentence 5 ′ shown in FIG. Then, the dependency analysis unit 101 performs dependency analysis on the text converted by the text shaping unit 105 to obtain a tree structure as illustrated in FIG. By the conversion on the grammar side by the sentence shaping unit 105, the dependency structure of the sentence 5 can be made equivalent to that of the sentence 7, and the accuracy of extracting synonymous expressions can be improved.

  The conversion rule shown in the present embodiment is an example, and the sentence shaping unit 105 increases the accuracy of synonymous expression extraction as the conversion rule having a similar sentence structure by dependency analysis is prepared in advance. be able to.

Embodiment 3 FIG.
Next, a third embodiment of the present invention will be described with reference to the drawings. FIG. 26 is a block diagram illustrating a configuration example of the synonymous expression extraction system 10 according to the third embodiment. As shown in FIG. 26, in the present embodiment, in addition to the components shown in FIG. 1, the similarity table database 106, the similarity reference correction means 107, and the input means 503 are included. And different.

  Specifically, the similarity table database 106 is realized by a database device such as a magnetic disk device or an optical disk device. The similarity table database 106 stores the similarity of the common expression structure for each synonym expression candidate pair, the similarity of the difference expression structure, and the content of the correctness determination of the synonym expression input by the system user.

  Specifically, the similarity reference correction unit 107 is realized by a CPU of an information processing apparatus that operates according to a program. The similarity criterion correction unit 107 has a function of dynamically changing the criterion for determining the similarity of different expressions in similar sentence pairs.

  Specifically, the input unit 503 is realized by an input device such as a keyboard or a mouse.

  In the present embodiment, when the output unit 502 outputs the extracted synonym expression according to the instruction of the similarity determination unit 104, the system user can input whether or not the output result is a true synonym expression. In this manner, synonymous expressions are output (displayed).

  For example, the output unit 502 outputs (displays) the synonymous expression in an output mode as shown in FIG. In the example shown in FIG. 27, the output means 502 displays the synonymous expression candidate pair together with the similarity in the similar sentence pair. In FIG. 27, a “judgment” column 728 at the lower right represents a column for inputting whether or not the synonym expression is correct. For example, the user performs an input instruction on whether the output synonymous expression is correct or incorrect by performing an input operation of “◯” or “X” from the “determination” column 728 of the display screen shown in FIG.

  In the example shown in FIG. 27, a case is shown in which a form that can input ○ or × is prepared to express correctness. However, the form is not limited to that shown in the present embodiment, and correct synonym expression is used. A display screen including radio buttons may be output (displayed) so that selection can be made. In addition, for example, the correctness of the output result of the synonym expression may be input as characters from the display screen.

  The input unit 503 inputs the correct / incorrect determination contents for the difference expression pair according to the input operation of the system user, and outputs them to the synonym expression extraction system 10. In this embodiment, as the input unit 503, for example, an input device such as a keyboard or a mouse is used. In the example shown in FIG. 27, the input unit 503 inputs the information of “◯” or “X” as the determination content of the synonym expression and outputs it to the synonym expression extraction system 10. Note that the input unit 503 may evaluate the extraction result of the synonymous expression by inputting a numerical value or the like for the similarity according to the operation of the system user. Moreover, the input means 503 may input by a numerical evaluation when inputting a numerical value, and may input by a continuous value.

  The similarity table database 106 stores the similarity of the common expression structure, the similarity of the difference expression structure, and the correctness determination contents of the synonym expression input by the system user as a similarity table. . FIG. 28 is an explanatory diagram showing an example of a similarity table stored in the similarity table database 106. The similarity table is a table (table) that summarizes the similarity of the common expression structure, the similarity of the difference expression structure, and the determination information of the synonym expression input by the system user in the similar sentence pair.

  The similarity determination unit 104 sets the similarity of the common expression structure to the i-th synonym expression candidate pair as ai, and the similarity of the different expression structure as bi. These may be calculated using the equations (3) and (4) shown in the first embodiment.

  Based on the information stored in the similarity table database 106, the similarity reference correction unit 107 estimates the weighting coefficient w used in the similarity calculation formula (2) shown in the first embodiment, and estimates The weighting factor is output to the similarity determination unit 104. w is a value for determining which of the similarity (ai) of the common expression structure and the similarity (bi) of the different expression structure is to be emphasized.

  An example of a process in which the similarity reference correction unit 107 corrects the weighting coefficient w is shown below. When the determination content is given by binary values of ◯ and ×, the processing is performed with x as 0 and ◯ as 1. Further, when the determination content is given as numerical data such as step evaluation or continuous value, the processing is performed using values normalized to 0 to 1.

  The similarity reference correction unit 107 calculates, for example, the maximum value and the minimum value of the determination information as values obtained by standardizing the determination contents, and then applies the equation (3) and uses the equation (3) ′. do it.

(Standardized judgment value) = (judgment−minimum value) / (maximum value−minimum value) Expression (3) ′

  Note that a standardized determination value using the expression (3) ′ is ti.

  FIG. 29 is an explanatory diagram showing a table in which the determination contents in the similarity table shown in FIG. 28 are quantified. Note that n is the number of synonymous expression candidate pairs in FIG. In this case, the similarity criterion correcting unit 107 estimates w where the difference between ti and the similarity w × ai + (1−w) × bi of the synonym candidate pair calculated using Expression (2) is small. . Specifically, the similarity criterion correcting unit 107 minimizes E shown in the equation (5) when the difference between w × ai + (1−w) × bi and ti in the equation (2) is ei. Find w.

E = Σei × ei Formula (5)

  In equation (5), Σ is a symbol representing the sum when i is changed from 1 to n. Further, the text shaping unit 105 obtains the formula (6) by differentiating the formula (5) with respect to w and setting the differential formula to 0.

w = Σ (ti−bi) (ai−bi) / Σ (ai−bi) (ai−bi) Equation (6)

  The text shaping unit 105 obtains w that minimizes E using Expression (6).

  Hereinafter, the correction method of w will be described using a specific example. FIG. 30 is an explanatory diagram showing a table in which the determination contents in the similarity table including four synonymous expression candidate pairs are digitized. In FIG. 30, it is assumed that true synonymous expressions are “ABC function”, “XYZ function”, “PQR function”, and “ABC function”.

  In addition, when synonym expression extraction is performed using Expression (2), the weighting factor w is set to 0.8, and the similarity threshold for determining whether or not to use synonym expression is 0.8. Suppose that it is set. The “similarity” shown in FIG. 30 indicates the similarity of each synonym expression candidate calculated by the similarity determination unit 104 using equation (2). In the example shown in FIG. 30, the similarity determination unit 104 includes “ABC function” and “XYZ function”, “ABC function”, “visualization diagnosis”, and “PQR” corresponding to the similarity indicated by shading. “Function” and “ABC function” are extracted as synonymous expressions. That is, in the example shown in FIG. 30, the similarity determination unit 104 erroneously extracts a set of “ABC function” and “visualization diagnosis” as a synonymous expression.

  In the example shown in FIG. 30, the synonym expression extraction system 10 inputs the determination content for the extraction result of the synonym expression as shown in the rightmost column of FIG. 30 according to the input operation of the system user. Then, the similarity reference correction unit 107 calculates the correction value of the weighting coefficient w to be about 0.17 using Equation (6).

  Next, the similarity determination unit 104 recalculates the similarity of each synonym expression candidate using the corrected weight coefficient w = 0.17. FIG. 31 is an explanatory diagram showing the calculation result of the similarity of each synonym expression candidate when w = 0.17.

  By the above processing, the similarity can be averaged close to the determination value (ti). For example, in the example shown in FIG. 31, “ABC function”, “XYZ function”, “PQR function”, and “ABC function”. The value of the weighting factor w can be corrected so that only the image has a similarity degree equal to or greater than the threshold value. Therefore, according to the present embodiment, the value of the weighting factor w can be estimated according to the above-described processing, and therefore the accuracy of synonymous expression extraction can be increased.

  In addition, although the case where the weighting coefficient w is corrected using Expression (6) has been shown, the correction method shown in the present embodiment uses Expression (2) as a similarity determination criterion for extracting synonymous expressions. In this case, another method may be used as a method of correcting the weighting factor w. For example, when a calculation formula other than the formula (2) is used as the similarity determination standard, the similarity reference correction unit 107 may set a correction function according to the calculation formula.

  Further, the similarity reference correction unit 107 may gradually change the weighting coefficient w instead of obtaining the correction amount uniquely as shown in the equation (6). For example, when the current weighting factor is w and the newly estimated weighting factor is w_new, the similarity criterion correcting unit 107 uses a method such as obtaining the weighting factor to be set as (w + w_new) / 2. The value of w may be corrected.

  Further, the configuration of the synonymous expression extraction system 10 shown in the present embodiment may further include the text shaping unit 105 shown in the second embodiment. If comprised in that way, a synonymous expression can be extracted more accurately.

Embodiment 4 FIG.
The synonymous expression extraction system 10 shown in each of the above embodiments includes, for example, a text reference relation extraction system that specifies the structure of document transmission of an electronic document, an expression unification system that unifies synonymous expressions included in the electronic document, The present invention can be applied to a text transmission evaluation system that evaluates document transmission. Hereinafter, a text reference relation system, an expression unification system, and a document transmission evaluation system to which the synonymous expression extraction system 10 is applied will be described with reference to the drawings.

  First, a fourth embodiment of the present invention will be described with reference to the drawings. FIG. 33 is a block diagram showing an example of the configuration of an expression unification system 300 using the document reference relation extraction system 100 to which the synonymous expression extraction system 10 according to the present invention is applied. In this example, the synonym expression extraction system 10 according to the present invention is applied as the synonym expression extraction device 350. Generally, in the process of transmitting an electronic document created for information sharing to another person (for example, transferring it using e-mail, etc.), the person who referred to the electronic document modifies the expression, thereby synonymous expression There is a tendency for electronic documents (parallel corpus) to be generated. In the present embodiment, using such a tendency, the document reference relationship extraction system 100 and the expression unification system 300 identify the hierarchical structure of document transmission and unify synonymous expressions between transmitted electronic documents. Perform the process.

  As shown in FIG. 33, the expression unification system 300 includes an organization information database 310, a transmission document extraction unit 320, a transmission document database 330, a document reference relation extraction system 100, a parallel corpus extraction unit 340, and synonymous expressions. An extraction device 350 and a synonym expression database 360 are included. The document reference relationship extraction system 100 includes a communication plan database 110, a non-reference relationship extraction unit 120, and a hierarchical structure specifying unit 130.

  In the present embodiment, the expression unification system 300 inputs various information from the input means 510 such as a keyboard and a mouse in accordance with a user operation. In the present embodiment, the expression unification system 300 collects electronic documents transmitted between users from the document providing server 520. In the present embodiment, the expression unification system 300 outputs (for example, displays) various information to the output unit 530 such as a display device.

  In this embodiment, the case where the input unit 510, the document providing server 520, and the output unit 530 are provided outside the expression unification system 300 is shown. , And output means 530. Further, the expression unification system 300 may have the function of the document providing server 520.

  In this embodiment, the case where the communication plan database 110 is provided inside the document reference relationship extraction system 100 is shown, but the communication plan database 110 may be provided outside the document reference relationship extraction system 100.

  In the present embodiment, the document reference relationship extraction system 100 and the expression unification system 300 are realized by, for example, a computer (for example, an information processing apparatus such as a personal computer) that operates according to a program.

  In the following description, an example will be described in which a member in a project submits “weekly report” as an electronic document in accordance with a communication plan using e-mail (e-mail). However, this example is for facilitating the description of the present invention and does not limit the scope of application of the present invention. Hereinafter, transferring an electronic document between users using a transfer means such as an electronic mail is expressed as transmitting the electronic document. An electronic document transmitted between users is also expressed as a transmitted document.

  In the present embodiment, the document reference relationship extraction system 100 is actually performed when an electronic document is transmitted in an organization such as a company according to a predetermined electronic document transmission plan (communication plan). Used to extract the hierarchical structure of electronic document transmission. The expression unification system 300 is used for the purpose of unifying synonymous expressions included between electronic documents before and after transmission using the hierarchical structure of electronic document transmission extracted by the document reference relation extraction system 100.

  Specifically, the input unit 510 is an input device such as a keyboard or a mouse. The input unit 510 inputs organization information in accordance with a user operation. The expression unification system 300 stores the organization information input from the input unit 510 in the organization information database 310. Moreover, the input means 510 inputs a communication plan according to a user's operation. The document reference relationship extraction system 100 stores the communication plan input from the input unit 510 in the communication plan database 110.

  Specifically, the organization information database 310 is realized by a database device such as a magnetic disk device or an optical disk device. The organization information database 310 stores organization information in advance. “Organization information” is information indicating the configuration of an organization where electronic documents are transmitted. In the present embodiment, the organization information includes a member ID for identifying each member involved in document transmission and a group ID for identifying the group to which each member belongs. The member ID is assigned to each member to identify each person, and is, for example, an employee number in a company. The group ID is an ID for identifying the belonging group. The organization information includes contact information such as an e-mail address for notifying the analysis result, personal information of each member such as name, age, job title, and specialized field as other information.

  In the present embodiment, it is assumed that the organization information is created in advance by an administrator in the organization and registered in the organization information database 310. For example, the organization information may be extracted from data recorded in a project or organization management tool, or may be input by a person using the input unit 510. In addition, as long as it is machine-readable, organization information written on a sheet or the like may be read by OCR or the like.

  FIG. 34 is an explanatory diagram of an example of organization information. According to the organization information shown in FIG. 34, it can be seen that the organization to which the expression unification system 300 is applied has seven members and three groups. One member may belong to a plurality of groups, or may serve as a plurality of positions. One member may have a plurality of e-mail addresses. For simplicity of explanation, in this embodiment, each member is assumed to belong to only one group.

  Specifically, the communication plan database 110 is realized by a database device such as a magnetic disk device or an optical disk device. The communication plan database 110 stores communication plans in advance. “Communication plan” is information indicating a transmission plan of an electronic document in an organization. The communication plan conveys “what document (document name)”, “when (scheduled date and time of transmission)”, “from whom (Fm)”, “to whom (To)”, and “how (transmission means)”. It includes plan information indicating whether or not to perform, and a plan number (No) for identifying each plan. A communication plan is created for a series of document transmissions. In this embodiment, it is assumed that one plan number (No) is given to one series of transmission plans. Note that these pieces of information are examples, and the information included in the communication plan is not limited to these.

  In this embodiment, it is assumed that the communication plan is created in advance by an administrator in the organization and registered in the communication plan database 110. For example, the communication plan is input from the input unit 510 in accordance with an operation of an administrator or the like. Also, for example, by using existing information extraction technology, information indicating the scheduled transmission date and time of the electronic document, the document name, the transmission source, the transmission destination, and the transmission means is extracted from a predetermined material (electronic document), and the communication plan The table may be automatically generated.

  FIG. 35 is an explanatory diagram showing an example of a communication plan. In the present embodiment, the same plan number (No) is assigned to a plan for transmitting one document having the same sender. In the communication plan shown in FIG. 35, the scheduled transmission date / time stores the scheduled date / time for transmitting each electronic document. The designation of the scheduled transmission date / time is not limited to a day unit, and may be performed in a unit of time or may include information such as morning or afternoon. Further, it may be specified by information indicating a period from when to when, or the scheduled transmission time of the electronic document may be specified using a linguistic expression such as every Friday or the end of the month.

  Note that each record of the communication plan is stored in the communication plan database 110 in the order of the scheduled transmission date and time. In the document name column, information for specifying an electronic document such as a subject name and a file name included in the e-mail is stored.

  Each member shall attach an e-mail subject name, file name, etc. in accordance with the document name. For example, when it is determined in advance that a document name of “weekly report” is assigned to a weekly electronic document, each member in the organization transmits an electronic document using an electronic mail. The subject of the email is “weekly report” and the electronic document is attached to the email and sent. Each member in the organization gives the file name “weekly report” to the created or edited electronic document.

  In FIG. 35, “Fm (transmission source)” and “To (transmission destination)” include “from whom” and “who” in the document transmission (that is, the transmission destination and transmission source of the transmission document). The member ID etc. which an individual can identify are used for description contents. However, information other than the member ID may be used as long as the information can identify each member. In addition, Fm and To can be specified by a group ID in order to reduce the burden of inputting a communication plan by a system user. When Fm is specified in a group and each member of the group creates and transmits a different electronic document, a different plan number (No) is given to each electronic document in accordance with the rules described above. .

  Further, in FIG. 35, a means for transmission is included in the transmission means column. For example, the communication plan transmission means run includes information indicating means such as e-mail (e-mail). The transmission means column is not limited to e-mail, but includes various transmission methods such as FAX (facsimile), mailing of printed materials, writing on an electronic bulletin board, and uploading of files to designated folders. May be. That is, any transmission means may be used as long as the electronic file of the transmitted electronic document can be specified.

  Communication plan data may be input by a person using the input unit 510, or may be read from a communication plan on paper by OCR or the like if machine-readable. In addition, when there is a plan of transmission by free description, a communication plan may be extracted from the plan using existing text processing technology.

  FIG. 35 shows an example of a communication plan related to “weekly report”, and 1 is assigned as a plan number (No). Further, according to each record shown in the plan No. (P001, P002, P003), on September 28, 2006, the members M3, M4, and M5 e-mail “weekly report” to each member of the group G2. It is understood that it is scheduled to be transmitted by (e-mail). Further, according to the record shown in the plan No. (P004), on the next day, September 29, 2006, the member M2 is scheduled to transmit “weekly report” to each member of the manager M1 and the group G2. I understand. Further, according to the record shown in the plan No. (P005), on September 29, 2006, the member M1 is scheduled to transmit “weekly report” to each member of the group 3 by e-mail. I understand.

  Specifically, the transfer document extraction unit 320 is realized by a CPU and a network interface unit of an information processing apparatus that operates according to a program. The transmission document extracting unit 320 transmits each electronic document from a document providing server 520 in which an electronic document created in a predetermined project or an electronic document transmitted using e-mail (e-mail) or the like is stored. A function for extracting document information is provided. In addition, the transmission document extracting unit 320 has a function of storing the extracted transmission document information in the transmission document database 330.

  In this embodiment, each electronic document is shared by using a shared server (document providing server 520) in the organization, and the document providing server 520 transmits log information (transmitted document information) to which each electronic document is transmitted. ) Is remembered. For example, the transmission document extraction unit 320 automatically accesses the document provision server 520 every predetermined period (for example, once every day) and receives the transmission document information from the document provision server 520 via a network such as a LAN. It is stored in the transmission document database 330. Further, for example, the transmission document extracting unit 320 automatically accesses the document provision server 520 using a new electronic document stored in the document provision server 520 as a trigger, and transmits the document from the document provision server 520 via the network. Information is received and stored in the transmission document database 330.

  Note that the document providing server 520 (shared server) is not limited to one, and a plurality of documents may be provided in the organization. For example, as the document providing server 520, a plurality of types of servers such as a mail server and a file server may be provided. Further, for example, as the document providing server 520, a plurality of common servers (for example, file servers) of the same type may be provided.

  The transmission document information is log information indicating the transmission status of the electronic document. In the present embodiment, the transmission document information includes “document ID”, “transmission date / time”, “document name”, “Fm (transmission source)”, “To (transmission destination)”, and “transmission means”. ing. Further, the transmission document information may include the contents of the electronic document together with information indicating these transmission records.

  The document ID is identification information for identifying each electronic document. In the present embodiment, one document ID is assigned to electronic documents having the same contents transmitted by the same person at the same date and time. The transmission date and time is information indicating the date and time when the electronic document is transmitted. Fm and To are information indicating from whom to whom the transmission document is transmitted (transmission source and transmission destination of the electronic document). The transmission means is information indicating what kind of media is used for document transmission. For example, the transmission document information includes information on various types of transmission means such as e-mail (electronic mail), FAX (facsimile), and mailing of printed matter as media.

  For example, in the case of transmission by e-mail, the transmission document extraction unit 320 determines the Fm (transmission source) and To (transmission destination) of each mail based on the header of the email or log information stored in the mail server. And the subject of the e-mail is extracted as the document name. Then, the transmission document extracting unit 320 stores the transmission document information including the extracted transmission source, transmission destination, and document name in the transmission document database 330.

  In the case of transmitting an electronic document by uploading it to a shared folder or a document management tool or the like, the transmission document extraction means 320 uses the file name of the uploaded file as the document name and designates the uploader as Fm (transmission source ), The transmission document information may be extracted with the person who referred to the file as To (transmission destination) and the transmission date and time as the upload date and time. Even if the electronic file is printed and mailed / handed, the transmission document extracting means 320 determines that the transmission by mail has been made if there is a transmission record regarding document transmission before and after the mail in the transmission process. It can also be added to the transmission record.

  Specifically, the transfer document database 330 is realized by a database device such as a magnetic disk device or an optical disk device. The transmission document database 330 stores the transmission document information extracted by the transmission document extraction unit 320.

  Note that the transmission document database 330 stores the transmission document information so that it can be determined whether the transmission is planned in the communication plan or not. For example, the transmission document database 330 stores the transmission document information in a distinguishable manner by adding a flag indicating whether or not there is a plan to each transmission document information, or changing the storage destination or file of the transmission document depending on the presence or absence of the plan. To do. Note that the transmission document extraction unit 320 assumes that the correspondence between each electronic document and each communication plan is such that Fm (transmission source) and To (transmission destination) of the transmission document information and the communication plan match. The condition can be determined based on the similarity of electronic document names. In addition, the transmission document extraction unit 320 may use an absolute condition that the character strings in the electronic document name are completely matched, or the ratio of the common character string in the electronic document name in order to cope with the notation fluctuation. The degree of similarity may be obtained and determined by, for example.

  For example, it is assumed that the document names in the communication plan and the transmission document information are “weekly report 20060928” and “weekly report September 28, 2006”, respectively, and have the same Fm (transmission source) and To (transmission destination). In this case, since there are 10 characters in common among the 11 characters of the average number of characters in both document names, the character matching rate is 10 / 11≈0.91. If this character match rate is used as the similarity, and the similarity is determined to be similar when the character match rate is 0.8 or more, the transfer document extracting unit 320 performs similarity determination. The communication plan can be associated with the transmission document information. Further, the transmission document extracting means 320 determines that “2006-09-28” and “September 28, 2006” have the same meaning using not only the number of matching characters but also an ontology dictionary or the like. For example, a semantic similarity measure may be used.

  If the scheduled transmission date / time of the communication plan is different from the transmission date / time of the transmission document information, the transmission document extraction unit 320 may determine that the communication plan and the transmission document information even if the document names are similar. Can be determined to have no corresponding relationship. For example, if the document name is “weekly report” and the actual transmission date / time is more than 7 days later than the scheduled transmission date / time, the transmission document extraction unit 320 may determine that the weekly report is for another week. it can. In this case, the difference between the transmission date / time for determining the correspondence in the electronic document and the scheduled transmission date / time can be arbitrarily determined by the user of the system.

  The transmission document extraction unit 320 develops the communication plan when measuring (determining) the similarity when the Fm (transmission source) and To (transmission destination) of the communication plan are designated by a group. . “Development of communication plan” refers to making communication plan data a communication plan between members by referring to the relationship between groups and members of organization information. At this time, those in which Fm (transmission source) and To (transmission destination) are the same during deployment are excluded from the transmission plan.

  FIG. 36 is an explanatory diagram showing an example of the transfer document information. In FIG. 36, the plan number (No) in the figure represents the plan number (No) of the corresponding communication plan. In FIG. 36, for example, it can be seen that the weekly report of the document ID “D101” was transmitted from the member M3 to the members M2, M4, and M5 at 17:00 on September 28, 2006, respectively.

  FIG. 37 is an explanatory diagram showing an example in which the communication plan “No. P001” is developed. In the example shown in FIG. 37, a communication plan indicating document transmission from the member M3 to the group G2 is developed into a communication plan indicating document transmission from the member M3 to each member of the group G2. However, in the example shown in FIG. 37, since the member M3 is also one of the members of the group G2, the transfer document extracting unit 320 determines that the record ((3) from the expanded communication plan to the document transfer from the member M3 to the member M3 ( The shaded portion 705) shown in FIG. 37 is deleted.

  The document reference relationship extraction system 100 has a function of specifying the hierarchical structure of document transmission based on the transmission document information stored in the transmission document database 330 and the communication plan stored in the communication plan database 110. Further, the document reference relationship extraction system 100 has a function of outputting information indicating the identified document transmission hierarchical structure to the parallel corpus extraction unit 340.

  In the present embodiment, the document reference relationship extraction system 100 executes a document transmission hierarchical structure specifying process in accordance with an instruction operation of a user (a member of any group in the organization). For example, when the user clicks the “execute” button with a mouse on a predetermined input form displayed on a display device included in the information processing apparatus, the document reference relation extraction system 100 starts a document transmission hierarchical structure specifying process.

  FIG. 38 is an explanatory diagram schematically showing document transmission in the communication plan shown in FIG. 35 as a network. In FIG. 38, nodes represent members, and arrows between the nodes represent electronic document transmission relationships. As shown in FIG. 38, the document transmission path extends from the nodes of the members M3, M4, and M5 in a network shape. For this reason, the document transmission hierarchical structure among the members M3, M4, and M5 cannot be specified as it is. The document reference relationship extraction system 100 extracts a reference relationship between electronic documents from the state of a network-like document transmission structure as shown in FIG.

  Specifically, the non-reference relationship extracting unit 120 is realized by a CPU of an information processing apparatus that operates according to a program. The non-reference relationship extraction unit 120 has a function of extracting document transmission that does not cause a reference relationship between electronic documents. First, the non-reference relationship extraction unit 120 creates a plan table based on the communication plan. The plan table is a table in which the presence / absence of a document transmission plan at the same scheduled transmission date / time in a communication plan with the same plan number (No) is summarized in a table format.

  The non-reference relationship extraction unit 120 creates a plan table according to the following process. First, the non-reference relationship extraction unit 120 obtains the union of Fm (transmission source) and To (transmission destination) in the communication plan. Next, the non-reference relationship extraction unit 120 creates a square matrix in which a set of corresponding members is arranged in rows and columns for each same scheduled transmission date and time. Here, the row direction of the created square matrix represents Fm (transmission source) in document transmission, and the column direction represents To (transmission destination). Next, the non-reference relationship extraction unit 120 assigns 1 to a place where there is a document transmission plan between Fm (transmission source) and To (transmission destination) in the plan table, and a place where there is no document transmission plan. 0 is assigned to.

  By performing the above processing, the non-reference relationship extraction unit 120 creates a plan table. FIG. 39 is an explanatory diagram showing a general example of a plan table created by the non-reference relationship extraction unit 120. In the plan table shown in FIG. 39, the leftmost column indicates the member ID of Fm (transmission source) in document transmission, and the top row indicates the member ID of To (transmission destination). In addition, the t-th schedule is expressed as Pt, and the larger the value of t, the later the scheduled transmission date and time. In the plan table shown in FIG. 39, a circled portion 707 means that there is a document transmission plan from member M1 to member M7.

  FIG. 40 is an explanatory diagram showing a plan table and a non-reference relationship table actually created by the non-reference relationship extraction unit 120 based on the communication plan shown in FIG. FIG. 40A shows a plan table (P1) for the scheduled transmission date and time of September 28, 2006. FIG. 40B shows a plan table (P2) for the scheduled transmission date and time of September 28, 2006. In FIG. 40, 0 is omitted to make the plan table easier to see.

  The non-reference relationship extracting means 120 is configured such that the i-th member Mi and the j-th member Mj in the plan table are the same as the portion 708 surrounded by the square frame in the plan table P1 shown in FIG. When the electronic document is transmitted to the member M1, it is determined that there is no reference relationship between the member Mi and the member Mj. That is, in general, when electronic documents are exchanged between two members, there is a strong tendency that correction / editing of electronic documents is not performed only by referring to the electronic documents. Therefore, using such a tendency, the non-reference relationship extraction unit 120 determines that there is no reference relationship when electronic documents are exchanged between two members.

  Specifically, as shown in FIG. 40, the non-reference relationship extracting unit 120 sets Pt (i, j) and Pt (j, i) to 1 in the plan table Pt having the same scheduled transmission date and time. Judge the transmission plan as a non-reference relationship. Here, P (i, j) represents the transmission relationship from member Mi to member Mj in the plan table. Then, the non-reference relationship extraction unit 120 generates a non-reference relationship table that is a table indicating the obtained non-reference relationship. FIG. 40C shows a non-reference relationship obtained from the plan table shown in FIG.

  As shown in FIG. 40C, the non-reference relationship table includes a transmission plan number (No), Fm (transmission source), and To (transmission destination). This means that there is no reference relationship in document transmission corresponding to Fm (transmission source) and To (transmission destination) of the plan number (No) shown in the non-reference relationship table. That is, in the transmission document information shown in FIG. 36, the document transmission shown in the transmission document information whose line numbers are 2, 3, 5, 6, 8, and 9, respectively, can be determined as document transmission in which no reference relationship occurs. .

  The non-reference relationship extraction unit 120 may create a non-reference relationship table using the organization information. For example, the non-reference relationship extraction unit 120 may be configured such that there is no reference between the same positions based on the organization information, or there may be no reference relationship between the same groups. Further, the non-reference relationship extraction unit 120 may create a non-reference relationship table by combining a plurality of conditions such as a group and a job title. It may be possible to delete it.

  Specifically, the hierarchical structure specifying unit 130 is realized by a CPU of an information processing apparatus that operates according to a program. The hierarchical structure specifying means 130 is a function that specifies “document transmission hierarchical structure” in the same type of electronic document group based on the communication plan stored in the communication plan database 110 and the transfer document information stored in the transfer document database 330. Is provided. In this case, the hierarchical structure specifying unit 130 uses the non-reference relationship table created by the non-reference relationship extracting unit 120 to specify the hierarchical structure of document transmission.

  FIG. 41 is an explanatory diagram showing an example of an algorithm for processing in which the hierarchical structure specifying unit 130 specifies the hierarchical structure of document transmission. The operation of the hierarchical structure specifying unit 130 will be described below according to the algorithm shown in FIG. First, the hierarchical structure specifying unit 130 extracts transfer document information corresponding to each communication plan from the transfer document database 330 (step A1 in FIG. 41). Next, the hierarchical structure specifying unit 130 creates a transmission table based on the transmission document information (step A2 in FIG. 41). The transmission table is a table in which the presence / absence of document transmission is summarized in the form of a table for each identical transmission date / time in the transmission document information in which the same scheduled transmission date / time is planned in the communication plan.

  Note that, when a plurality of transfer tables are created, the hierarchical structure specifying unit 130 executes the process shown in FIG. 41 for all the transfer tables. Therefore, in step A2, the hierarchical structure specifying unit 130 substitutes 1 for a coefficient t indicating what number of transmission tables is being processed.

  The hierarchical structure specifying unit 130 creates a transmission table according to the following procedure. First, the hierarchical structure specifying unit 130 obtains the union of Fm (transmission source) and To (transmission destination) from the transmission document information corresponding to the designated communication plan. Next, the hierarchical structure specifying unit 130 creates a square matrix in which a set of corresponding members is arranged in rows and columns for each same transmission date and time. Here, the row direction of the created square matrix represents Fm (transmission source) in document transmission, and the column direction represents To (transmission destination). Next, the hierarchical structure specifying unit 130 assigns 1 to a place where there is a document transmission plan between Fm (transmission source) and To (transmission destination) in the plan table, and places it in a place where there is no document transmission plan. Gives 0.

  By performing the above processing, the hierarchical structure specifying unit 130 creates a transmission table. FIG. 42 is an explanatory diagram showing a general example of a transfer table created by the hierarchical structure specifying unit 130. In the transmission table shown in FIG. 42, the leftmost column indicates the member ID of Fm (transmission source) in document transmission, and the top row indicates the member ID of To (transmission destination). In the transmission table shown in FIG. 42, a circled portion 710 means that the document transmission from the member M5 to the member M7 has been performed. Hereinafter, the t-th transmission table is denoted by Ct. Also, the larger the value of t, the later the transmission date and time.

  FIG. 43 is an explanatory diagram showing five transmission tables actually created by the hierarchical structure specifying unit 130 based on the transmission document information shown in FIG. 36, and an example of specifying the hierarchical structure from the transmission table. Note that the transmission table shown in FIG. 43 is an excerpt of only the portion where document transmission occurred (1 exists in the line).

  Next, the hierarchical structure specifying unit 130 specifies the document transmission structure in each transmission table (step A3 in FIG. 41). In this case, the hierarchical structure specifying unit 130 can extract Fm (transmission source) and To (transmission destination) at a location where 1 is assigned in the transmission table Ct. The link means that a document is transmitted from a member of Fm (transmission source) to a member of To (transmission destination). For example, the link is transmitted from Fm (transmission source) to To (transmission). This corresponds to a process (for example, a process of assigning link information) for connecting arrows toward (destination).

  The hierarchical structure specifying unit 130 performs link using the link conditional expression (7). However, in this case, the hierarchical structure specifying unit 130 performs processing so as not to extract the non-reference relationship based on the non-reference relationship table.

  In equation (7), i represents the index of the member in the vertical direction in the transfer table Ct, and j represents the index of the member in the horizontal direction. Further, the consequent part (that is, the part after then) in the equation (7) means that the document transmission from the member Mi to the member Mj is performed. In FIG. 43, the schematic diagram shown on the right side of each transmission table is a hierarchical structure of document transmission obtained by the hierarchical structure specifying unit 130 by applying Expression (7) to the transmission table shown in FIG.

  Through the above processing, in the present embodiment, the hierarchical structure specifying unit 130 first creates a transmission table including both the reference relationship and the non-reference relationship, and further creates the transmission table and the non-reference relationship table. Based on this, a hierarchical structure of document transmission that does not include a non-reference relationship is specified. The hierarchical structure specifying unit 130 may create a transmission table that does not include a non-reference relationship in advance based on the non-reference relationship table. By doing so, the data capacity of the transfer table created by the hierarchical structure specifying means 130 can be reduced.

  Next, the hierarchical structure specifying unit 130 extracts the start and end of document transmission from each transmission table (step A4 in FIG. 41). Here, it is assumed that the document transfer initiator in the transfer table Ct is Ms_t and the document transfer end is Me_t. In this case, the hierarchical structure specifying unit 130 can extract the initiator Ms_t and the endor Me_t using the equations (8) and (9), respectively.

  In Expressions (8) and (9), Nm indicates the number of members, and Ma_t indicates the a-th member in the transfer table Ct. As shown in FIG. 43, in each transfer table, there may be a plurality of document transfer starters and endorsers.

  Next, the hierarchical structure specifying unit 130 applies the processing of step A3 and step A4 to all transfer tables, and then specifies the hierarchical structure of document transfer between the transfer tables. Specifically, when the start and end of document transmission are specified, the hierarchical structure specifying means 130 is whether or not processing has been completed for all transmission tables (t = Nτ (total number of transmission tables)). (Step A5). If not t = Nτ, the hierarchical structure specifying unit 130 adds 1 to the value of t (step A6), returns to the process of step A3, and repeatedly executes the processes of steps A3 and A4. If t = Nτ, the hierarchical structure specifying unit 130 specifies the structure of document transmission between the transmission tables (step A7). In this case, the hierarchical structure specifying unit 130 specifies the document transmission structure by specifying the end of document transmission in the transmission table Cτ performing document transmission to the transmission starter in the transmission table Ct. be able to. However, it is assumed that t> τ.

  Next, the hierarchical structure specifying unit 130 can specify the document transfer structure between transfer tables by linking the initiator to the specified endorser. In this case, the hierarchical structure specifying unit 130 specifies the document transmission structure using the link conditional expression (10).

  FIG. 44 is an explanatory diagram showing the hierarchical structure of document transmission specified by the hierarchical structure specifying unit 130 from the transmission document information shown in FIG. 36 in accordance with the processing shown in FIG. In FIG. 44, each node 712a represents each member, and each arrow 712b represents a document transmission relationship. Also, the square frame shown in FIG. 44 shows the document ID of the electronic document transmitted by each adjacent node.

  Note that the processing for calculating the reference relationship of document transmission includes To (transmission destination) and Fm (transmission source) of different paths in a set of one-way paths of Fm (transmission source) and To (transmission destination). It is positioned as a technology that connects and creates directed grabs under constraints. The restriction in the process of calculating the document transfer reference relationship is that To (transmission destination) and Fm (transmission source) are the same in the connection, and Fm (transmission source) is later than To (transmission destination). That is. That is, if this constraint is satisfied and a directed graph can be extracted, the hierarchical structure specifying unit 130 may use a reference relationship extraction method other than the algorithm shown in FIG. 41 when calculating the document transfer reference relationship.

  For example, the hierarchical structure specifying unit 130 extracts document transmission end persons and members who transmit electronic documents to the document transmission end persons, and further extracts members transmitting electronic documents to the members. The document transfer reference relationship may be calculated using a method of recursively repeating the above process. In this case, the hierarchical structure specifying unit 130 ends the process when the corresponding member does not exist or when the document transmission having a non-reference relationship is extracted as the end condition. By executing such processing, the hierarchical structure specifying unit 130 creates a tree structure of transmission with the end of the document transmission as the root, the document transmission initiator as the leaf, and the document transmission relay as the intermediate node. Can be identified.

  FIG. 45 is an explanatory diagram showing the process of tracing from the member M6, one of the document transfer endorsers, to the document transfer starter in the transfer document information shown in FIG. In FIG. 45, a blank area indicates non-reference-related document transmission. The hierarchical structure specifying unit 130 performs the same processing for each transmission end person (for example, member M6 and member M7), and finally regards a common part in the tree structure of the document transmission related to each end person as one, and merges them. Thus, the hierarchical structure similar to the document transmission hierarchical structure shown in FIG. 44 is specified. The hierarchical structure may be corrected / input / deleted by a person using the input means 510.

  By executing the processing shown in FIG. 45, the hierarchical structure specifying unit 130 can eliminate the need to create a transfer table when specifying the hierarchical structure of document transmission, and specify the hierarchical structure of document transmission. It is possible to reduce the data capacity of the data created at the time.

  As shown in this embodiment, if a transmission table is created when specifying the hierarchical structure of document transmission, the hierarchical structure of repeated document transmission can be specified later by reusing the transmission table. can do. Further, the hierarchical structure of document transmission can be analyzed and edited by performing editing work using the created transmission table at a later date.

  Specifically, the parallel corpus extraction unit 340 is realized by a CPU of an information processing apparatus that operates according to a program. The parallel corpus extraction unit 340 has a function of extracting a parallel corpus based on the document transmission hierarchical structure identification result output from the hierarchical structure identification unit 130.

  In general, a relayer in a hierarchical structure of document transmission tends to modify an electronic document to be transmitted. For this reason, electronic document pairs before and after a relayer in the document transmission process tend to be a parallel corpus. Therefore, in the present embodiment, the parallel corpus extraction means 340 extracts a parallel corpus using such a tendency generally found in document transmission.

  For example, in the example shown in FIG. 44, the parallel corpus extraction unit 340 has a parallel corpus between the electronic document D101, D102 or D103 and the electronic document D104 before and after document transmission to the member M2, because the member M2 is a relay person. Judge that there is. For example, the parallel corpus extraction unit 340 determines that the electronic document D104 and the electronic document D105 before and after document transmission to the member M1 are parallel corpuses because the member M1 is a relay person.

  In general, when document transmission is performed in an extremely short period (for example, the same day), there is a strong tendency that the electronic document to be transmitted is transferred to the next member without modification or editing. Therefore, the parallel corpus extraction means 340 considers such a tendency and executes a process for determining that the contents of the electronic document are the same in a transmission plan having the same scheduled transmission date and time in the communication plan. May be. For example, the parallel corpus extraction means 340 determines that the member M2 is merely a bridging member for the electronic document because the transmission from the member M2 to the member M1 is planned on the same day. The content may be determined. Note that the similar sentence pair database 501 used in the first to third embodiments may record a sentence including many similar words as a similar sentence in the parallel corpus obtained by the parallel corpus extraction unit 340. .

  Specifically, the synonym expression extraction device 350 is realized by a CPU of an information processing device that operates according to a program. The synonym expression extraction device 350 has a function of extracting synonym expressions from the text information of the parallel corpus specified by the parallel corpus extraction unit 340. The synonym expression extraction device 350 has a function of storing the extracted synonym expressions in the synonym expression database 360. The synonym expression extraction device 350 has a function of causing the output unit 530 to output the extracted synonym expression.

  The synonym expression extraction device 350 extracts synonym expressions from the parallel corpus using various existing extraction techniques as synonym expression extraction techniques. For example, the synonym expression extraction device 350 applies dependency analysis to similar sentence pairs as a technique for extracting synonym expressions from a parallel corpus, and uses a method of extracting a subtree satisfying a predetermined condition as a synonym expression. Also good. In this case, for example, the synonym expression extraction device 350 includes, as predetermined conditions, (1) the root of the subtree obtained by dependency analysis is a predicate, and (2) the paired subtrees include a common specific expression. (3) A synonym expression may be extracted using three cases that a subtree includes a case required by each precaution.

  The technique for extracting the synonymous expression shown above is, for example, the document “Shinyama Y, and Sekine S,“ Paraphrase acquisition for information extraction ”, 2nd International Workshop on Paraphrasing: Paraphrase Acquisition and Applications, pp. 65-71, 2003 ".

  Specifically, the synonymous expression database 360 is realized by a database device such as a magnetic disk device or an optical disk device. The synonym expression database 360 stores the synonym expressions extracted by the synonym expression extraction device 350. FIG. 46 is an explanatory diagram showing an example of a storage format of synonym expressions stored in the synonym expression database 360. As shown in FIG. 46, the synonym expression database 360 associates a synonym expression ID and a synonym expression pair, the document ID of the electronic document in which these synonym expressions are included, and the member who created the electronic document. Remember. In this case, the synonym expression database 360 stores synonym expressions so that the transfer functions (“Fm (transmission source)” and “To (transmission destination)”) of the electronic document can be understood.

  In FIG. 46, the synonym expression ID is identification information given to uniquely identify a pair of synonym expressions. As shown in FIG. 46, in the synonym expression ID 700 (W001), it is understood that the expression XYZ has been changed to the synonym expression ABC in the electronic document ID 104 created by referring to the electronic document D101 created by the member M3. . That is, the pair of the expression XYZ and the expression ABC is synonymous with each other.

  The storage format shown in FIG. 46 is an example of a storage method of synonymous expressions, and the synonymous expression database 360 stores synonymous expressions in other storage formats as long as the information of synonymous expressions can be extracted. Also good.

  Specifically, the output means 530 is realized by a display device such as a display device. For example, when the output unit 530 is a display device, the extraction result (the extracted synonym expression) of the synonym expression extraction device 350 is displayed according to the instruction of the synonym expression extraction device 350. Instead of displaying the extraction result of the synonym expression, the expression unification system 300 may output the extraction result of the synonym expression as an electronic file. Further, the expression unification system 300 may notify (send) the extraction result to each member's terminal using e-mail or the like using the address information indicated in the organization information.

  FIG. 47 is an explanatory diagram showing an example of the extraction result of the synonym expression output by the output unit 530. As shown in FIG. 47, the output means 530 extracts the synonym expression in an output format that shows which synonym expression is used between which electronic documents that have been communicated between whom. Output the result. In addition, the display mode when the output unit 530 outputs the extraction result of the synonymous expression is not limited to the display mode illustrated in FIG. 47, and may be output in another display mode.

  The example shown in FIG. 47 indicates that two sets of synonymous expressions (ABC and XYZ, PQR and STU) are used for the electronic document D101 and the electronic document D104. In addition, a pair of synonymous expressions (LMN and STU) are used for the electronic document D102 and the electronic document D104. Further, as shown in FIG. 44, the output unit 530 may display the hierarchical structure of document transmission and may superimpose the used synonymous expressions.

  In the present embodiment, the storage device (not shown) of the information processing apparatus that implements the document reference relation extraction system 100 and the expression unification system 300 stores various programs for extracting reference relations between electronic documents. I remember it. For example, the storage device of the information processing apparatus stores a document reference relationship extraction program for causing a computer to execute a hierarchical structure specifying process for specifying a hierarchical structure of document transmission for an electronic document transmitted between members.

  In the present embodiment, the storage device of the information processing apparatus that implements the document reference relationship extraction system 100 and the expression unification system 300 extracts a parallel corpus that is an electronic document that includes synonymous expressions, and converts the parallel corpus into a parallel corpus. Various programs for unifying synonymous expressions included are stored. For example, the storage device of the information processing apparatus extracts non-reference relationship information indicating document transmission that does not have a reference relationship between electronic documents based on transmission plan information indicating a transmission plan of electronic documents stored in a database in advance in a computer. Non-reference relationship extraction processing and hierarchical structure specifying processing for specifying the hierarchical structure of document transmission for electronic documents transmitted between members, and transmission records of electronic documents stored in the database are stored in the hierarchical structure specifying processing. Based on the transmission record information shown, a process for identifying a transmission record corresponding to the transmission plan indicated in the transmission plan information stored in the database in advance, and from the initiator of the electronic document transmission to the end person based on the identified transmission record Based on the processing that identifies the hierarchical structure up to and the extracted non-reference relationship information, the hierarchical structure from the start to the end of the transmission of the electronic document is not referenced. Stores representations unified program for executing a process of identifying without the document transmission is related.

  Next, the overall operation of the expression unification system 300 will be described. FIG. 48 is a flowchart showing an example of processing for specifying the hierarchical structure of document transmission and unifying the expression of the transmission document.

  First, the transfer document extraction unit 320 of the expression unification system 300 extracts transfer document information from the document providing server 520 at a predetermined timing and stores it in the transfer document database 330 (step S11). For example, the transfer document extracting unit 320 automatically accesses the document providing server 520 every predetermined period (for example, once every day), receives the transfer document information from the document providing server 520, and stores it in the transfer document database 330. Further, for example, the transmission document extraction unit 320 automatically accesses the document provision server 520 and receives the transmission document information from the document provision server 520, triggered by a new electronic document stored in the document provision server 520. It is stored in the transmission document database 330.

  By repeatedly executing the process of step S11 at a predetermined timing, the transfer document information is stored in the transfer document database 330 and updated as needed.

  Next, when an instruction operation is performed by the user (a member of any group in the organization), the document reference relation extraction system 100 starts a process for specifying a hierarchical structure of document transmission. In the process of specifying the hierarchical structure of document transmission, the non-reference relationship extraction unit 120 of the document reference relationship extraction system 100 firstly, based on the organization information stored in the organization information database 310 and the communication plan stored in the communication plan database 110. A plan table is created and a non-reference relationship table is created (step S12).

  Next, the hierarchical structure specifying unit 130 creates a transmission table based on the communication plan stored in the communication plan database 110 and the transfer document information stored in the transfer document database 330 (step S13).

  Next, the hierarchical structure specifying unit 130 specifies the hierarchical structure of document transmission based on the created transmission table (step S14). In this case, the hierarchical structure specifying unit 130 uses the non-reference relationship table created by the non-reference relationship extracting unit 120 to specify a hierarchical structure excluding document transmission that does not cause a reference relationship between electronic documents.

  Next, the parallel corpus extracting unit 340 extracts a parallel corpus based on the document transmission hierarchical structure specified by the hierarchical structure specifying unit 130 (step S15). Next, the synonym expression extraction device 350 extracts the synonym expression from the transmission document based on the parallel corpus extracted by the parallel corpus extraction unit 340, and stores the extraction result in the synonym expression database 360 (step S16). Also, the synonym expression extraction device 350 causes the output unit 530 to output the extraction result of the synonym expression.

  As described above, according to the present embodiment, the document reference relationship extraction system 100 identifies a document transmission hierarchical structure in which a reference relationship is likely to occur between electronic documents based on organization information, a communication plan, and transmission document information. To do. Therefore, when the electronic document is transmitted, the reference relationship of the electronic document can be extracted.

  Further, according to the present embodiment, it is possible to extract a reference relationship between electronic documents, and the expression unification system 300 refers to another person's transmission document in the transmission process, and relays document transmission for aggregation and distribution. A person is efficiently discovered and a parallel corpus is extracted. Therefore, a parallel corpus between transmission documents can be extracted efficiently.

Embodiment 5 FIG.
Next, a fifth embodiment of the present invention will be described with reference to the drawings. FIG. 49 is a block diagram illustrating a configuration example of the expression unification system 300 using the document reference relationship extraction system 100 according to the fifth embodiment. As shown in FIG. 49, this embodiment differs from the fourth embodiment in that it includes a hierarchical structure database 370, a correction plan input means 540, and a correction means 380 in addition to the components shown in FIG. Different.

  Specifically, the hierarchical database 370 is realized by a database device such as a magnetic disk device or an optical disk device. The hierarchical structure database 370 stores the hierarchical structure specified by the document reference relationship extraction system 100.

  Specifically, the correction proposal input means 540 is realized by an input device such as a keyboard or a mouse. The correction plan input means 540 has a function of inputting a correction plan for the extracted synonym expression according to a user operation.

  Specifically, the correcting unit 380 is realized by a CPU and a network interface unit of an information processing apparatus that operates according to a program. The correction unit 380 has a function of correcting the synonym expression based on the correction plan input from the correction plan input unit 540. Further, the correcting means 380 has a function of specifying a member that notifies that the synonymous expression has been corrected.

  Next, the operation will be described. FIG. 50 is a flowchart illustrating an example of processing for correcting a synonymous expression. In the present embodiment, the document reference relationship extraction system 100 specifies the hierarchical structure of document transmission according to the same processing as in the fourth embodiment. Further, the expression unification system 300 extracts synonymous expressions between transfer documents according to the same processing as in the fourth embodiment. Hereinafter, the operation for correcting the synonymous expression will be described with reference to the flowchart shown in FIG. In the present embodiment, the case where the output unit 530 is a display device will be described as an example.

  First, the expression unification system 300 displays a list of synonym expressions stored in the synonym expression database 360, for example, on a display device that is the output unit 530 (step B1 in FIG. 50). In addition, the expression unification system 300 displays a correction input screen for correcting the synonymous expression on the display device. For example, the expression unification system 300 displays the correction input screen shown in FIG. In the example shown in FIG. 51, the synonym expression ID, the extracted synonym expression, and the document ID of the electronic document that is the extraction source are displayed on the correction input screen, and an input for inputting a correction plan for each synonym expression. A column 717 is provided.

  Next, the correction plan input means 540 inputs a correction plan for unifying expressions according to the operation of the system user or member (step B2 in FIG. 50). For example, an input device such as a keyboard or a mouse is used as the correction proposal input unit 540. In FIG. 49, the correction plan input unit 540 and the input unit 510 are shown separately, but the correction plan input unit 540 and the input unit 510 may be realized by the same input device.

  Note that the correction proposal may be input in a manner of selecting any of the used synonymous expressions, or a new expression may be input in a free description format. The synonymous expression can be modified by a user or member of any system.

  The expression unification system 300 inputs the correction plan and the synonym expression ID input to the synonym expression from the correction plan input means 540 according to the operation of the system user or member.

  Next, the correcting unit 380 extracts, from the document providing server 520, the electronic document having the document ID including the synonym expression to be corrected based on the transfer document information stored in the transfer document database 330. In this case, for example, the correcting unit 380 specifies the document ID of the electronic document including the synonymous expression to be corrected based on the transmitted document information. Further, the correcting unit 380 transmits an electronic document extraction request corresponding to the specified document ID to the document providing server 520 via the network. Then, the document providing server 520 extracts the electronic document in response to the request and transmits it to the expression unifying system 300 via the network.

  Next, based on the electronic document extracted from the document providing server 520, the correction unit 380 causes the output unit 530 to output a sentence before and after the document ID and its synonymous expression as a correction candidate (step B3 in FIG. 50). In this case, the correction unit 380 causes the output unit 530 to display a sentence having an arbitrary length as a sentence before and after the synonymous expression. For example, the correcting unit 380 may display the preceding and following words including synonymous expressions. In addition, the correction unit 380 may display the sentence unit or may not display the correction candidates.

  The correction unit 380 uses the output unit 530 to display a display screen as shown in FIG. 52, for example, to output correction candidates. In the example shown in FIG. 52, the correction means 380 displays a display screen including the document ID of the electronic document to be corrected, the date and time when the electronic document was transmitted, the member using the synonymous expression, and the correction candidate sentence. ing. The display screen shown in FIG. 52 further includes an input field 718 for inputting whether or not to agree with the correction.

  Next, the amendment proposal input means 540 inputs that the amendment is agreed among the amendment candidates according to the operation of the system user or member, and the expression unification system 300 agrees to the amendment from the amendment proposal input means 540. The instruction information to the effect is input (step B4 in FIG. 50).

  In addition, in order to input that it is agreement or non-agreement, in the example shown in FIG. 52, the case where it inputs using the symbol of (circle) or x is shown. However, the input method does not need to be a selection format using symbol input such as ◯ and X as long as it can be notified to the expression unification system 300 that the correction is agreed. For example, it may be possible to input using a radio button, or to input agreement or non-agreement by inputting characters using a keyboard or the like. In addition to inputting information indicating consent or disagreement, in FIG. 52, a column for inputting information indicating a reason for correction or a reason for not correction may be provided on the display screen.

  Next, the correcting unit 380 replaces the synonymous expression instructed to agree with the correction, and unifies the expressions between the transmission documents (step B5 in FIG. 50). Note that the correction unit 380 performs replacement processing on the synonym expression using, for example, an existing text processing technique.

  Next, the correction means 380 notifies the system user and member of which expression of the transmitted document that has been transmitted becomes a synonymous expression and how it has been corrected (step B6 in FIG. 50).

  In step B6, the correcting unit 380 specifies a hierarchical structure including the document ID of the corrected electronic document from the hierarchical structure database 370, and extracts all document IDs included in the specified hierarchical structure. Next, the correcting unit 380 searches the transfer document database 330 using the extracted document ID as a search key, and specifies transfer document information from the transfer document database 330. The correcting unit 380 specifies an electronic document corresponding to the specified transmission document information from the electronic documents extracted from the document providing server 520. Then, the correcting unit 380 specifies an electronic document including a synonymous expression pair (ABC or XYZ in the example illustrated in FIG. 51) in the specified electronic document.

  Next, the correcting unit 380 extracts members who have played the role of Fm (transmission source) or To (transmission destination) for the identified electronic document. That is, the member who played the role of transmission source or transmission destination is involved in document transmission including synonymous expressions, and refers to the transmission document. Then, the correction unit 380 notifies the member involved in document transmission regarding the correction. For example, the correction unit 380 transmits a message indicating that the correction for unifying synonymous expressions has been performed and the correction location to the terminal of the member involved in document transmission via the network.

  By executing the above processing, a notification regarding correction of the synonym expression is given to the minimum necessary members related to the reference of the synonym expression.

  For example, according to the transmission document information shown in FIG. 36, the member M4 is transmitted the electronic document D101 and the electronic document D104 from the member M3 and the member M2, respectively. In addition, the expression ABC of the electronic document D101 and the expression XYZ of the electronic document D104 have a synonymous expression relationship. However, since the member M4 does not know the transmission process of the electronic document and the generation process of synonymous expressions, it is difficult to determine whether or not the expressions are the same.

  In the present embodiment, the correction unit 380 notifies (transmits) the message shown in FIG. 53 to the member M4, for example. The correcting unit 380 may transmit the message shown in FIG. 53 to the member terminal using e-mail (e-mail), for example. By receiving the message shown in FIG. 53 and confirming the content of the displayed message, the member M4 can grasp the cause of the synonymous expression, the correction process, and the like.

  As described above, according to the present embodiment, the hierarchical structure of document transmission related to the electronic document from which synonymous expressions are extracted is specified based on the specification result of the hierarchical structure of document transmission stored in the hierarchical structure database 370. To do. In addition, it is possible to specify an electronic document that has been transmitted in the hierarchical structure of document transmission. Furthermore, based on the transmission document information stored in the transmission document database 330, the Fm (transmission source) and To (transmission destination) of the transmission document can be specified. Therefore, it is possible to notify the members involved in the corrected electronic document that the synonym expression has been corrected.

  For example, in the conventional systems, apparatuses, methods, and programs described in Patent Documents 1 to 3, when an electronic document is modified, people who refer to the electronic document cannot be specified. For example, in the systems, apparatuses, methods, and programs described in Patent Document 1 and Patent Document 3, there is no processing for dealing with a case where a reference document is modified.

  Further, for example, in the apparatus and method described in Patent Document 2, the notification of document correction is limited to the corrector of the electronic document and the owner of the electronic document referenced from the corrected document, and the reference relationship is also HTML or the like. It is a hyperlink in a standard format. Therefore, it is impossible to notify people related to the free-form electronic document before / after correction that the correction has been made. Note that the “related people” are an electronic document corrector to be notified by the apparatus and method described in Patent Document 2, an owner of the electronic document referenced from the corrected document, and the like. In addition, “related people” refers to those who have been transferred to or viewed the electronic document, those who have created the electronic document by referring to the revised document, and who have viewed the created electronic document. is there.

  On the other hand, in the present embodiment, as described above, the reference relationship of the electronic document can be extracted, and which information is transmitted to whom based on the contents of the electronic document and the transmission structure. Can be determined. Therefore, it is possible to specify a member who should be notified of a change in the contents of an electronic document.

Embodiment 6 FIG.
Next, a sixth embodiment of the present invention will be described with reference to the drawings. FIG. 54 is a block diagram illustrating a configuration example of an expression unification system 300 using the document reference relationship extraction system 100 according to the sixth embodiment. As shown in FIG. 54, this embodiment is different from the fourth embodiment in that it includes a trend analysis means 390 in addition to the components shown in FIG. Further, in this embodiment, the function of the hierarchical structure specifying unit 130 is different from the function of the hierarchical structure specifying unit 130 shown in the fourth embodiment.

  Specifically, the trend analysis unit 390 is realized by a CPU of an information processing apparatus that operates according to a program. The trend analysis means 390 has a function of analyzing the usage trend of the synonymous expression of each member.

  Further, in the present embodiment, the hierarchical structure specifying unit 130, in addition to the functions shown in the fourth embodiment, is based on a communication plan stored in the communication plan database 110, and a document transmission hierarchy planned in advance. Has the function to specify the structure. In the present embodiment, the hierarchical structure specifying unit 130 also has a function of specifying the hierarchical structure of document transmission actually performed according to the same processing as in the fourth embodiment.

  Next, the operation will be described. In the present embodiment, the expression unification system 300 specifies the hierarchical structure of document transmission actually performed according to the same processing as in the fourth embodiment, and executes the processing for unifying the expression of the transmission document. . Further, in this embodiment, the expression unification system 300 identifies the hierarchical structure of document transmission planned in advance and presents the occurrence tendency of synonymous expressions in addition to the processing shown in the fourth embodiment. Execute the process. FIG. 55 is a flowchart showing an example of processing for specifying a planned document transmission hierarchical structure and presenting the occurrence tendency of synonymous expressions.

  The non-reference relationship extraction unit 120 creates a plan table and a non-reference relationship table for any communication plan stored in the communication plan database 110 according to the same processing as in the fourth embodiment (step S31). ). FIG. 56 is an explanatory diagram showing an example of a process in which the non-reference relationship extraction unit 120 identifies two plan tables created based on the communication plan shown in FIG. 35 and a hierarchical structure of document transmission from the plan table. In the plan table shown in FIG. 56, only the portion where the document transmission plan is present (1 exists in the line) is extracted and shown.

  In the present embodiment, the non-reference relationship extraction unit 120 creates a non-reference relationship table similar to the non-reference relationship table shown in FIG. 40 based on the plan table shown in FIG. 56, as in the fourth embodiment. The description will be made assuming that it has been extracted. The non-reference relationship extraction unit 120 develops the communication plan based on the organization information stored in the organization information database 310 when the group ID is included in the communication plan.

  Next, the hierarchical structure specifying unit 130 is preliminarily planned based on the communication plan by applying an algorithm similar to the algorithm applied to the transfer table in the fourth embodiment to the plan table. Identify the document transmission hierarchy. In this case, the hierarchical structure specifying unit 130 specifies a planned document transmission hierarchical structure based on the plan table and the non-reference relationship table created by the non-reference relationship extraction unit 120 (step S32).

  Next, the hierarchical structure specifying unit 130 specifies the document transmission structure in each plan table. The hierarchical structure specifying unit 130 links the document transmission structure in each of the plan tables with Fm (transmission source) and To (transmission destination) at a location where 1 is assigned in the t-th plan table Pt. Can be specified by. In this case, the hierarchical structure specifying unit 130 links the transmission source and the transmission destination using the link conditional expression (11). However, the hierarchical structure specifying unit 130 performs processing so as not to extract the transmission source and the transmission destination that are in the non-reference relationship based on the non-reference relationship table. The link here has the same meaning as the link used in the fourth embodiment.

  In Expression (11), i is the index of the member in the vertical direction in the plan table Pt, and j is the index in the horizontal direction. Further, the consequent part in the equation (11) means that the document transmission from the member Mi to the member Mj is planned. 56, the schematic diagram shown on the right side of each plan table is a hierarchical structure specified by the hierarchical structure specifying unit 130 by applying the formula (11) to the plan table shown in FIG.

  Next, the hierarchical structure specifying unit 130 extracts a document transmission start scheduled person and a terminal planned person from each plan table. Here, it is assumed that the person scheduled to start document transmission in the plan table Pt is Ms_t, and the person scheduled to end document transmission is Me_t. In this case, the hierarchical structure specifying unit 130 can extract the scheduled starter Ms_t and the scheduled ender Me_t using the equations (12) and (13), respectively.

  In Expression (12) and Expression (13), Nm indicates the number of members, and Ma_t indicates the a-th member in the plan table Pt. As shown in FIG. 56, in each plan table, there may be a plurality of document transmission start scheduled persons and terminal scheduled persons.

  Next, the hierarchical structure specifying unit 130 specifies the document transmission hierarchical structure between the plan tables after performing the same processing as steps S31 and S32 on all the plan tables. In this case, the hierarchical structure specifying means 130 extracts the end planner of the transmission of the plan table Pτ, which is a plan for performing the document transmission, with respect to the person scheduled to start the transmission in the plan table Pt. The hierarchical structure of transmission can be specified. However, it is assumed that t> τ.

  Next, the hierarchical structure specifying unit 130 specifies the document transmission structure between the planning tables by linking the scheduled starter with the extracted scheduled endorser. In this case, the hierarchical structure specifying unit 130 specifies the document transmission structure between the planning tables using the link conditional expression (14).

  By executing the above processing, the hierarchical structure specifying unit 130 specifies the hierarchical structure shown in FIG. 57 based on the plan table shown in FIG. In FIG. 57, each node 722a represents a member, and each arrow 722b represents a document transmission plan. In addition, in the square frame 722c shown in FIG. 57, the plan number (No) of the communication plan corresponding to each adjacent arrow is shown.

  As described above, as shown in FIG. 57, this system can identify a hierarchical structure of document transmission planned in advance, and can extract a reference relationship between electronic documents scheduled to be transmitted based on a plan table.

  As in the case where the document reference relationship extraction system 100 shown in the fourth embodiment extracts the reference relationship based on the transmitted document information, the method for extracting the reference relationship from the communication plan is shown in this embodiment. The method is not limited.

  For example, the hierarchical structure specifying unit 130 extracts a document transmission end candidate, extracts a member who transmits an electronic document to the document transmission end candidate, and further transmits the electronic document to the member. A preliminarily planned reference relationship of document transmission may be calculated using a method of recursively repeating the process of extracting members. In this case, the hierarchical structure specifying unit 130 ends the process when the corresponding member does not exist or when the document transmission having a non-reference relationship is extracted as the end condition. By executing such processing, the hierarchical structure specifying means 130 uses the document transfer terminal planner as the root, the document transfer start planner as the leaf, and the document transfer relay planner as the intermediate node. A tree structure can be specified.

  In addition, the hierarchical structure specifying unit 130 performs the same processing for each terminal candidate (for example, member M6 and member M7) of each transmission, and finally, there is one common part in the tree structure of document transmission related to each terminal candidate. By regarding and merging, a hierarchical structure similar to the document transmission hierarchical structure shown in FIG. 57 is specified.

  Next, the trend analysis means 390 first counts how often each member used the synonym expression with which member based on the extraction result of the synonym expression stored in the synonym expression database 360, A synonymous expression usage count table that collectively shows the number of times of use of synonymous expressions for each member is generated (step S33). Then, the trend analysis unit 390 causes the output unit 530 to output the generated synonym expression usage count table (step S34).

  The use of the synonym expression means that the electronic document is transmitted using an expression different from the expression in the transmitted electronic document, that is, the synonym expression is generated.

  There are various methods for counting synonymous expressions. For example, the trend analysis unit 390 may count the number of electronic documents that contain synonymous expressions. In addition, the trend analysis unit 390 may count the number of sentences including synonymous expressions, or may count the number of synonymous expressions.

  FIG. 58 is an explanatory diagram showing an example of an output format of the synonym expression usage count table. In the example shown in FIG. 58, a synonym expression usage count table among seven members is shown, and synonyms when an electronic document is transmitted from the member shown in the leftmost column to the member shown in the top row. Indicates the number of expressions used. Further, by obtaining the sum of the count values in the vertical direction in the synonym expression use count table shown in FIG. 58, the number of use times of the synonym expression used by one member for the electronic document transmitted from the other member can be obtained. it can. Further, by obtaining the sum of the count values in the horizontal direction, it is possible to obtain the number of times that the synonymous expression used by another member is used for an electronic document of a certain member.

  The output unit 530 displays a synonymous expression usage count table so that the synonymous expression usage tendency of each member can be understood in accordance with an instruction from the trend analyzing unit 390, for example, in the case of a display device or the like. Note that the output unit 530 may display the data of the synonymous expression usage count table collectively according to the instruction of the trend analysis unit 390, or may display the data of the synonymous expression usage count table as a graph.

  Also, the trend analysis means 390 superimposes the identified document transmission hierarchical structure and the number of synonymous expressions used, for example, as shown in FIG. 59, in order to show the number of times of synonymous expressions used easily for the user of the system. You may display on the output means 530. FIG. In FIG. 59, each node 724a indicates a member, and an arrow 724b between the nodes indicates a document transmission plan. In FIG. 59, a square frame 724c on the arrow indicates the number of synonymous expressions used between corresponding members.

  By displaying in the display format shown in FIG. 59, it is possible to easily grasp to what extent a synonymous expression may be used in future document transmission. Therefore, the user of the system can request a member who uses many synonymous expressions to present a numerical rationale and reduce the use of synonymous expressions. In addition, as a numerical value given to the synonym expression usage count table, for example, a ratio between a transmitted electronic document and an electronic document using the synonym expression may be used instead of using only the number of synonym expressions used.

  As described above, according to the present embodiment, it is possible to specify in advance a hierarchical structure of document transmission planned in advance based on a communication plan input in advance before document transmission. Therefore, it is possible to extract the electronic document reference relationship in future document transmission.

  For example, the conventional systems, apparatuses, methods, and programs described in Patent Documents 1 to 3 cannot extract reference relationships between electronic documents to be transmitted in the future. In the systems, devices, methods, and programs described in Patent Documents 1 to 3, the generated electronic information such as the frequency of rare words appearing between electronic documents, the category of electronic documents, the reference time / frequency of electronic documents, and reference links The reference relationship is extracted based on the information extracted from the document. Therefore, even when electronic documents to be created are determined in the future, it is not possible to extract a reference relationship that occurs between these electronic documents.

  In contrast, in the present embodiment, as described above, the hierarchical structure of document transmission is specified based on the communication plan, so that the reference relationship can be extracted without using the electronic document transmission record. Therefore, the reference relationship can be extracted from the future transmission plan.

  Further, according to the present embodiment, based on the organization information and the communication plan, it is possible to extract a preliminarily planned reference relationship of the transmission document, and based on the number of times the synonym expression is used between past members. It is possible to quantify the likelihood of synonymous expressions. Therefore, it is possible to specify in advance a document transmission that is likely to generate a synonymous expression and become a parallel corpus in the future.

  In the present embodiment, the document reference relationship extraction system 100 and the expression unification system 300 specify the hierarchical structure of the actual document transmission, execute the process of unifying the expression of the transmission document, The case where the hierarchical structure of the planned document transmission is identified and the process of presenting the occurrence tendency of the synonymous expression is executed was shown, but the generation structure of the synonymous expression is presented by identifying the hierarchical structure of the document transmission planned in advance Only the processing to be performed may be executed. In this case, the expression unification system 300 may not include the transmission document extraction unit 320 and the transmission document database 330.

  In addition, it identifies the hierarchical structure of the actual document transmission and executes a process that unifies the expression of the transmitted document, identifies the hierarchical structure of the planned document transmission, and presents the occurrence tendency of synonymous expressions. You may make it comprise the system which performs the process to perform as a separate system.

Embodiment 7 FIG.
Next, a seventh embodiment of the present invention will be described with reference to the drawings. In the present embodiment, a case will be described in which the document reference relationship extraction system 100 is applied to a document transmission evaluation system that evaluates actual document transmission. FIG. 60 is a block diagram illustrating a configuration example of a document transmission evaluation system 400 using the document reference relationship extraction system 100 according to the seventh embodiment. As shown in FIG. 60, the present embodiment is different from the fourth embodiment in that it does not include the parallel corpus extraction means 340 and the synonym expression extraction means 350 among the components shown in FIG. Further, the present embodiment is different from the fourth embodiment in that a difference extracting means 550 is included in addition to the components shown in FIG.

  Specifically, the difference extraction unit 550 is realized by a CPU of an information processing apparatus that operates according to a program. The difference extraction unit 550 compares the communication plan stored in the communication plan database 110 with the transfer document information stored in the transfer document database 330 and clarifies whether or not the document transfer according to the communication plan is performed. Is provided.

  The difference extraction unit 550 clarifies the difference between the transmission document information and the communication plan, and causes the output unit 530 to output a comparison result between the transmission document information and the communication plan. The difference extraction unit 550 extracts differences between the following three items for the transmission document information and the communication plan.

  First, the difference extraction means 550 extracts document transmissions that are planned but not transmitted in the communication plan (difference 1: document not transmitted). Further, the difference extracting unit 550 extracts the document transmission that is being transmitted even though it is not planned in the communication plan (difference 2: transmission unplanned). Also, the difference extraction means 550 extracts the document transmission that has been transmitted as planned in the communication plan, but at a time different from the planned transmission time (difference 3: transmission date / time difference).

  First, the difference extraction unit 550 extracts a communication plan that cannot be matched with the transmission document information from the communication plan database 110. That is, the difference extraction unit 550 extracts a communication plan including document transmission that has not been transmitted although document transmission is planned. Specifically, the difference extraction unit 550 extracts, from the communication plan database 110, a communication plan that includes information on the scheduled transmission date and time that does not match the information on each transmission date and time indicated in the transmission document information stored in the transmission document database 330. . By executing this processing, the difference extraction unit 550 extracts the difference 1 (document not transmitted).

  Next, all electronic documents corresponding to the same document ID as the document ID included in the hierarchical structure of document transmission specified by the hierarchical structure specifying unit 130 are specified based on the transfer document information stored in the transfer document database 330. To do. In this case, the difference extraction unit 550 also identifies transmission document information that does not correspond to the communication plan. That is, the difference extraction unit 550 extracts the transmission document information including the document transmission that is being transmitted even though the document transmission is not planned. Specifically, the difference extraction unit 550 extracts from the transmission document database 330 transmission document information including information on transmission date and time that does not match the information on each scheduled transmission date and time indicated in the communication plan stored in the communication plan database 110. . By executing this processing, the difference extraction unit 550 extracts the difference 2 (transmission unplanned).

  Finally, the difference extraction unit 550 calculates the difference between the scheduled transmission date and time in the communication plan and the transmission date and time in the transmission document information corresponding to the communication plan. In this case, the difference extraction unit 550 calculates a difference between the transmission date and the scheduled transmission date by subtracting the transmission date and time (the date and time when the document transmission was actually performed) from the scheduled transmission date and time as a difference calculation method. If the calculation result is a negative value, it means that it has been transmitted by the scheduled transmission date and time. Further, if the calculation result is a positive value, it means that the transmission is delayed after the scheduled transmission date and time. If the scheduled transmission date and time of the communication plan is specified in a period having a certain range, the difference extraction unit 550 transmits the transmission date and time and the transmission when the electronic document is transmitted within the period. What is necessary is just to obtain | require the difference with a scheduled date and time as 0.

  Next, the difference extraction unit 550 creates the difference between the communication plan and the transmission document information as a difference table as shown in FIG. 61, for example, and causes the output unit 530 to output the created difference table. In the example shown in FIG. 61, the difference table includes, in each row, a plan ID based on a communication plan and a scheduled transmission date / time, a transmission document ID / transmission date / time based on transmission document information, Fm (transmission source), To (transmission destination), and a transmission schedule. Information on the difference between the date and time and the transmission date is included.

  In the difference table shown in FIG. 61, document transmission indicated by a row in which the transmission document ID is blank means document transmission that is planned but not transmitted (difference 1: no document transmission). In addition, document transmission indicated by a line having a blank plan ID means transmitted document transmission (difference 2: transmission unplanned) even though it is not planned. Further, as shown in FIG. 61, in the document transmission indicated by the line including both the transmission document ID and the plan ID, the result of calculating the difference between the transmission scheduled date and time and the transmission date and time is shown. And the calculation result of the difference between the transmission date and time corresponds to (difference 3: transmission date and time difference).

  In the example shown in FIG. 61, the difference between the scheduled transmission date and the transmission date and time is calculated on a date basis. However, the difference extraction means 550 is not limited to calculating the transmission date / time difference in units of days, and may calculate the transmission date / time difference in units of time, for example.

  In the example shown in FIG. 61, for example, the member M3 transmits the electronic document with the document ID 101 to the member M8, but it is understood that this document transmission is not planned in the communication plan. Also, it can be seen that the member M4 has failed to communicate to the member M2 planned in the communication plan P002. In addition, it can be seen that the member M5 is transmitting the electronic document one day later than the scheduled transmission date and time planned in the communication plan P003. In this way, the difference table can be used to express three differences between the communication plan and the transmission document information.

  For example, in the case of a display device, the output unit 530 outputs (displays) information of the difference table generated by the difference extraction unit 550 in accordance with an instruction from the difference extraction unit 550. For example, the output unit 530 may display the display screen shown in FIG. 62 in accordance with the instruction from the difference extraction unit 550.

  As shown in FIG. 62, in accordance with an instruction from the difference extraction unit 550, the output unit 530 first displays members included in the hierarchical structure of document transmission from the communication plan data as a solid line node 727a. Further, the output unit 530 is a member who has not been planned but has been transmitted according to the instruction of the difference extraction unit 550 (a member (To (transmission destination) in a row where the plan number (No) is blank in the difference table)). Is displayed as a broken-line node 727c.

  Further, the output unit 530 refers to the transmission document information in accordance with the instruction from the difference extraction unit 550 and displays the nodes in a display form in which the nodes are connected with arrows so that the hierarchical structure relationship can be understood. In other words, the output unit 530 displays the non-reference relationship between the nodes in a form that is not connected with an arrow even if document transmission is performed. In this case, the output unit 530 displays the display mode connected to the node indicated by the broken line by the solid line arrow 727c.

  Further, the output unit 530 displays in a display mode in which nodes that are not actually transmitted with an electronic document are connected by broken-line arrows 727d although document transmission is planned according to the communication plan according to the instruction of the difference extraction unit 550. To do. Even in this case, the output unit 530 displays the nodes having the same role in a manner that does not connect the arrows.

  Further, the output unit 530 displays the transmission date and time difference 727e enclosed by a square frame in the direction indicated by the solid line in accordance with the instruction from the difference extraction unit 550. The output means 530 displays the information in the difference table by displaying in the display mode as shown in FIG. 62, and clearly displays the difference between the communication plan and the actual document transmission.

  As shown in FIG. 62, it is understood that the document transmission from the member M4 to the member M2 and the document transmission from the member M1 to the member M7 are planned in advance but are not actually performed. Further, it can be seen that the document transmission from the member M3 to the member M8 and the document transmission from the member M2 to the member M9 are performed even though they are not planned in advance in the communication plan.

  Further, although the document transmission from the member M2 to the member M1 is delayed by one day, when the transmission path to the member M1 is referred, the transmission document from the member M4 has not arrived, and the document transmission from the member M4 You can see that is already one day late. From this, it can be understood that the cause of the document transmission delay was a problem in the document transmission of the members M3 and M4 rather than the transmission to the member M1. That is, from the situation shown on the display screen shown in FIG. 62, it can be easily determined that there is a problem in document transmission of members M3 and M4 rather than member M1.

  In the present embodiment, the storage device of the information processing apparatus that implements the document reference relation extraction system 100 and the document transmission evaluation system 400 stores various programs for evaluating the transmission status of electronic documents. For example, the storage device of the information processing apparatus stores the electronic record on the computer based on the transmission record information indicating the transmission record of the electronic document stored in the database and the transmission plan information indicating the transmission plan of the electronic document stored in the database in advance. A document transmission evaluation program for executing a difference extraction process for extracting a difference between a document transmission record and a transmission plan, and a difference output process for outputting a difference between the extracted electronic document transmission record and the transmission plan is stored. ing.

  As described above, according to the present embodiment, the difference extracting unit 550 compares the document transmission information with the hierarchical structure of the document transmission schedule specified by the hierarchical structure specifying unit 130 based on the communication plan. Further, the difference extraction means 550 is a document transmission that is included in the communication plan but has no transmission record in the transmission document information, a document transmission that is not included in the communication plan, or an electronic document transmission. The difference between the scheduled date and time and the transmission date and time of the electronic document corresponding to the plan is extracted as a difference in document transmission. Then, the difference extraction unit 550 causes the output unit 530 to display the extracted document transmission difference. Therefore, it is possible to clearly evaluate the person who referred to the electronic document and the appropriateness of the timing of the electronic document transmission.

  The present invention is suitably applied to an information search system that executes synonymous expression search and a program for realizing the information search system. The present invention is preferably applied to a project management system, a document management system, a project management system, and a program for realizing a document management system that perform terminology control and the like.

It is a block diagram which shows an example of a structure of the synonymous expression extraction system by this invention. It is a flowchart which shows an example of a synonym expression extraction system 10 extraction process of a synonym expression. It is explanatory drawing which shows the example of the similar sentence pair which a similar sentence pair database memorize | stores. It is explanatory drawing which shows the example of the analysis result of dependency analysis. It is explanatory drawing which shows the example of the data format of the information of the relationship between the clause and morpheme which a dependency analysis means memorize | stores in a recording medium. It is explanatory drawing which shows the example of a dependency table. It is explanatory drawing which shows the example of a definition of structural relationship. It is explanatory drawing which shows the example of a structural relationship specific table. It is explanatory drawing which shows the production | generation method of a structural relationship specific table. It is explanatory drawing which shows the example of the structural relationship specified with respect to the sentence 1. FIG. It is explanatory drawing which shows the example of the structural relationship specified with respect to the text 2. FIG. It is explanatory drawing which shows the example of a comparison structure relationship. It is explanatory drawing which shows the example of the comparison structure relationship in the text 1 and the text 2. FIG. It is explanatory drawing which shows the example of the identity determination table | surface of the structural relationship in a similar sentence pair. It is explanatory drawing which shows the other example of a similar sentence pair. It is explanatory drawing which shows the dependency structure with respect to the text 3 and the text 4. FIG. It is explanatory drawing which shows the example of the identity of the structural relationship in a similar sentence pair. It is explanatory drawing which shows the example of the identity of the structural relationship in a similar sentence pair. It is explanatory drawing which shows the example of the similarity calculation using the weighted average which the similarity determination means 104 calculated | required. It is explanatory drawing which shows the example of an output of synonymous expression. It is explanatory drawing which shows the other output example of synonymous expression. It is explanatory drawing which shows the principle of synonymous expression extraction in this invention. It is a block diagram which shows the structural example of the synonymous expression extraction system in 2nd Embodiment. It is explanatory drawing which shows the example of text shaping. It is explanatory drawing which shows the example of the dependency structure before and behind text shaping. It is a block diagram which shows the structural example of the synonymous expression extraction system in 3rd Embodiment. It is explanatory drawing which shows the example of an output which can input the correctness of the extracted synonymous expression. It is explanatory drawing which shows the similarity table used in order to adjust the similarity standard which the similarity table database 106 memorize | stores. It is explanatory drawing which shows the similarity table which digitized the determination content of the synonymous expression right / wrong by the system user shown in FIG. It is explanatory drawing which shows the relationship between the weighting coefficient of the expression of synonymous expression extraction, a threshold value, and similarity. It is explanatory drawing which shows the relationship between the weighting coefficient after correction | amendment of the expression of synonymous expression extraction, a threshold value, and similarity. It is explanatory drawing which shows the principle of synonymous expression extraction in a prior art. It is a block diagram which shows an example of a structure of the expression unification system using the document reference relationship extraction system to which the synonymous expression extraction system by this invention is applied. It is explanatory drawing which shows an example of organization information. It is explanatory drawing which shows an example of a communication plan. It is explanatory drawing which shows an example of transmission document information. It is explanatory drawing which shows the example which expand | deployed the communication plan. FIG. 36 is an explanatory diagram schematically showing document transmission in the communication plan shown in FIG. 35 as a network. It is explanatory drawing which shows the general example of the plan table which a non-reference relationship extraction means produces. It is explanatory drawing which shows the plan table and non-reference relationship table which the non-reference relationship extraction means actually created based on the communication plan shown in FIG. It is explanatory drawing which shows an example of the algorithm of the process in which a hierarchical structure specification means specifies the hierarchical structure of document transmission. It is explanatory drawing which shows the general example of the transmission table which a hierarchical structure specification means produces. It is explanatory drawing which shows the example which specifies a hierarchical structure from five transmission tables which the hierarchical structure specification means actually created based on the transmission document information shown in FIG. 36, and a transmission table. FIG. 43 is an explanatory diagram showing a hierarchical structure of document transmission specified by the hierarchical structure specifying unit from the transmission document information shown in FIG. 36 in accordance with the processing shown in FIG. FIG. 37 is an explanatory diagram showing a process of performing a trace from a member M6, which is one of the document transfer endorsers, to the document transfer starter in the transfer document information shown in FIG. It is explanatory drawing which shows an example of the memory | storage format of a synonym expression which a synonym expression database memorize | stores. It is explanatory drawing which shows an example of the extraction result of the synonymous expression which an output means outputs. It is a flowchart which shows an example of the process which specifies the hierarchical structure of document transmission and unifies the expression of a transmission document. It is a block diagram which shows the structural example of the expression unification system using the document reference relationship extraction system in 5th Embodiment. It is a flowchart which shows an example of the process which corrects synonymous expression. It is explanatory drawing which shows the example of the form (display screen) which inputs the correction proposal of synonymous expression. It is explanatory drawing which shows the example of the form (display screen) which inputs the display of a correction candidate, and consent to correction. It is explanatory drawing which shows the example of the display method (message) which notifies that correction of synonymous expression was performed. It is a block diagram which shows the structural example of the expression unification system using the document reference relationship extraction system in 6th Embodiment. It is a flowchart which shows an example of the process which identifies the hierarchical structure of document transmission planned beforehand and shows the generation | occurrence | production tendency of synonymous expression. It is explanatory drawing which shows the example of the process in which the non-reference relationship extraction means specifies the hierarchical structure of document transmission from two plan tables created based on the communication plan shown in FIG. 35, and a plan table. It is explanatory drawing which shows the example of the hierarchical structure of the document transmission identified from the communication plan. It is explanatory drawing which shows the example of the output format of a synonymous expression use frequency table. It is explanatory drawing which shows the example which output the usage count of synonymous expression with the hierarchical structure of document transmission. It is a block diagram which shows the structural example of the document transmission evaluation system using the document reference relationship extraction system in 7th Embodiment. It is explanatory drawing which shows the example of a difference table. It is explanatory drawing which shows the example which output the difference between the communication plan of a transmission document, and transmission document information.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 Synonymous expression extraction system 101 Dependency analysis means 102 Expression comparison means 103 Structural relationship specification means 104 Similarity determination means 105 Text shaping means 106 Similarity table database 107 Similarity reference correction means 501 Similar sentence pair database 502 Output means 503 Input means

Claims (16)

A synonym expression extraction system that extracts synonym expressions from pairs of similar sentences that are pairs of sentences that are similar to each other,
A structural relationship between common expressions that are expressions included in each sentence of the similar sentence pair and a structural relation between a difference expression that is included only in each sentence of the similar sentence pair and a common expression are specified. Structural relationship identification means;
Based on the structural relationship specified by the structural relationship specifying means, the difference expression included in the similar sentence pair as a synonym candidate, and a similarity calculation means for obtaining the similarity of the synonym candidate,
The common expression group included in common in each sentence of the similar sentence pair is in a similar relative positional relationship in the sentence structure of each sentence, and the difference expression group that is different between the sentences of the similar sentence pair A synonym expression extracting means for extracting a synonym expression from the pair of similar sentences based on the fact that the common expression group is in a similar relative positional relationship in the sentence structure of each sentence ;
The synonym expression extraction unit is configured to determine whether or not to extract a candidate for the synonym expression as a synonym expression based on the similarity obtained by the similarity calculation unit . A synonym expression extraction system that extracts synonym expressions from pairs of similar sentences that are pairs of sentences that are similar to each other,
Sentence structure specifying means for specifying the sentence structure of the pair of similar sentences;
Based on the sentence structure specified by the sentence structure specifying means, a common expression that is an expression commonly included in each sentence of the similar sentence pair and a difference that is an expression included only in each sentence of the similar sentence pair An expression extraction means for extracting an expression;
Based on the common expression and the difference expression extracted by the expression extraction means, the relative position relationship between the common expressions in the sentence structure of each sentence of the similar sentence pair, and the relative position relationship between the common expression and the difference expression A positional relationship specifying means for specifying
Based on the relative positional relationship specified by the positional relationship specifying means, similarity calculating means for calculating the similarity of the difference expression in the similar sentence pair from the identity of the structural relationship;
The common expression group included in common in each sentence of the similar sentence pair is in a similar relative positional relationship in the sentence structure of each sentence, and the difference expression group that is different between the sentences of the similar sentence pair A synonym expression extracting means for extracting a synonym expression from the pair of similar sentences based on the fact that the common expression group is in a similar relative positional relationship in the sentence structure of each sentence;
The synonym expression extraction unit extracts a synonym expression based on the similarity calculated by the similarity calculation unit.
Synonymous expression extraction system characterized by that. A synonym expression extraction system that extracts synonym expressions from pairs of similar sentences that are pairs of sentences that are similar to each other,
Sentence structure specifying means for specifying the sentence structure of the pair of similar sentences;
Based on the sentence structure specified by the sentence structure specifying means, a common expression that is an expression commonly included in each sentence of the similar sentence pair and a difference that is an expression included only in each sentence of the similar sentence pair An expression extraction means for extracting an expression;
Based on the common expression and the difference expression extracted by the expression extraction means, a positional relationship specifying means for specifying a relative positional relationship between the common expression and the difference expression in the sentence structure of each sentence of the similar sentence pair;
Based on the relative positional relationship specified by the positional relationship specifying means, similarity calculating means for calculating the similarity of the difference expression in the similar sentence pair from the identity of the structural relationship;
The common expression group included in common in each sentence of the similar sentence pair is in a similar relative positional relationship in the sentence structure of each sentence, and the difference expression group that is different between the sentences of the similar sentence pair A synonym expression extracting means for extracting a synonym expression from the pair of similar sentences based on the fact that the common expression group is in a similar relative positional relationship in the sentence structure of each sentence;
The synonym expression extraction unit extracts a synonym expression based on the similarity calculated by the similarity calculation unit.
Synonymous expression extraction system characterized by that.   The synonymous expression extracting means extracts the synonymous expression based on the fact that the different expression group and the common expression group are similar in the sentence structure of each sentence in the similar sentence pair. Item 4. The synonymous expression extraction system according to any one of Items 3 to 3. Any one of claims 1 to 4 further comprising a sentence conversion means for converting a sentence included in a similar sentence pair using a predetermined sentence conversion rule so that the sentence structure of the similar sentence becomes a similar structure. The synonymous expression extraction system according to claim 1. The synonym expression extraction means extracts the synonym expression from the pair of similar sentences according to a predetermined similarity criterion for extracting the synonym expression,
A similarity criterion correcting unit that corrects a similarity criterion for extracting the synonym expression based on correct / incorrect information indicating the correctness of the extraction result of the synonym expression extracted by the synonym expression extracting unit and the similarity of the synonym expression. The synonymous expression extraction system according to any one of claims 1 to 5 . A synonym expression extraction method for extracting synonym expressions from a pair of similar sentences in which sentences similar to each other are paired,
The computer has a structural relationship between common expressions that are expressions included in each sentence of the similar sentence pair, and a structural relationship between a difference expression and a common expression that is an expression included only in each sentence of the similar sentence pair. A structural relationship identifying step for identifying
Based on the identified structural relationship, the computer uses a difference expression included in the similar sentence pair as a synonym expression candidate, and calculates a similarity degree of the synonym expression candidate,
The computer has a common relative expression group included in each sentence of the similar sentence pair in a similar relative positional relationship in the sentence structure of each sentence, and a difference that differs between the sentences of the similar sentence pair A synonym expression extracting step of extracting a synonym expression from the pair of similar sentences based on the fact that the expression group and the common expression group are in a similar relative positional relationship in the sentence structure of each sentence ,
The synonym expression extraction method, wherein the computer determines whether or not to extract a candidate for the synonym expression as a synonym expression based on the obtained similarity in the synonym expression extraction step . A synonym expression extraction method for extracting synonym expressions from a pair of similar sentences in which sentences similar to each other are paired,
A sentence structure specifying step for specifying a sentence structure of the pair of similar sentences;
The computer, based on the specified sentence structure, a common expression that is an expression that is commonly included in each sentence of the similar sentence pair and a difference expression that is an expression that is included only in each sentence of the similar sentence pair An expression extraction step for extracting
Based on the extracted common expression and difference expression by the computer, the relative position relationship between the common expressions in the sentence structure of each sentence of the similar sentence pair, and the relative position relationship between the common expression and the difference expression A positional relationship identifying step for identifying
A similarity calculation step in which the computer calculates the similarity of the difference expression in the similar sentence pair from the identity of the structural relationship based on the identified relative positional relationship;
The computer includes a common expression group that is commonly included in each sentence of the similar sentence pair in a similar relative positional relationship in the sentence structure of each sentence, and a different expression that is different between the sentences of the similar sentence pair A synonym expression extracting step for extracting a synonym expression from the pair of similar sentences based on the fact that the group and the common expression group are in a similar relative positional relationship in the sentence structure of each sentence,
The computer extracts a synonym expression based on the calculated similarity in the synonym expression extraction step.
A synonymous expression extraction method characterized by that. A synonym expression extraction method for extracting synonym expressions from a pair of similar sentences in which sentences similar to each other are paired,
A sentence structure specifying step for specifying a sentence structure of the pair of similar sentences;
The computer, based on the specified sentence structure, a common expression that is an expression that is commonly included in each sentence of the similar sentence pair and a difference expression that is an expression that is included only in each sentence of the similar sentence pair An expression extraction step for extracting
A positional relationship specifying step in which the computer specifies a relative positional relationship between the common representation and the difference representation in the sentence structure of each sentence of the similar sentence pair based on the extracted common expression and difference expression;
A similarity calculation step in which the computer calculates the similarity of the difference expression in the similar sentence pair from the identity of the structural relationship based on the identified relative positional relationship;
The computer has a common relative expression group included in each sentence of the similar sentence pair in a similar relative positional relationship in the sentence structure of each sentence, and a difference that differs between the sentences of the similar sentence pair A synonym expression extracting step of extracting a synonym expression from the pair of similar sentences based on the fact that the expression group and the common expression group are in a similar relative positional relationship in the sentence structure of each sentence,
The computer extracts a synonym expression based on the calculated similarity in the synonym expression extraction step.
A synonymous expression extraction method characterized by that. Computer, so that the similar sentence structure similar sentence structure, using a predetermined text conversion rules, among the claims 7 to 9 including the text conversion step of converting the text contained in the similar sentence pairs The synonymous expression extraction method according to any one of the above. In the synonym expression extraction step, the computer extracts the synonym expression from the pair of similar sentences according to a predetermined similarity criterion for extracting the synonym expression,
The computer includes a similarity criterion correction step of correcting a similarity criterion for extracting the synonym expression based on correct / incorrect information indicating correctness of the extraction result of the extracted synonym expression and the similarity of the synonym expression.
The synonymous expression extraction method according to any one of claims 7 to 10 . A synonymous expression extraction program for extracting synonymous expressions from a pair of similar sentences in which similar sentences are paired,
On the computer,
A structural relationship between common expressions that are expressions included in each sentence of the similar sentence pair and a structural relation between a difference expression that is included only in each sentence of the similar sentence pair and a common expression are specified. Structural relationship identification processing,
Based on the identified structural relationship, the difference expression included in the pair of similar sentences is regarded as a synonym candidate, and a similarity calculation process for obtaining the similarity of the candidate synonym expression;
The common expression group included in common in each sentence of the similar sentence pair is in a similar relative positional relationship in the sentence structure of each sentence, and the difference expression group that is different between the sentences of the similar sentence pair Based on the fact that the common expression group is in a similar relative positional relationship in the sentence structure of each sentence, the synonym expression extracting process for extracting the synonym expression from the similar sentence pair is executed,
A synonym expression extraction program for executing a process of determining whether or not to extract a candidate for the synonym expression as a synonym expression based on the obtained similarity in the synonym expression extraction process . A synonymous expression extraction program for extracting synonymous expressions from a pair of similar sentences in which similar sentences are paired,
On the computer,
A sentence structure specifying process for specifying a sentence structure of the pair of similar sentences;
Based on the specified sentence structure, a common expression that is an expression that is commonly included in each sentence of the similar sentence pair and a difference expression that is an expression that is included only in each sentence of the similar sentence pair are extracted. Expression extraction processing,
Based on the extracted common expression and difference expression, a position for specifying a relative positional relationship between the common expressions in the sentence structure of each sentence of the similar sentence pair and a relative position relation between the common expression and the difference expressions Relationship identification processing;
Based on the identified relative positional relationship, a similarity calculation process for calculating the similarity of the difference expression in the similar sentence pair from the identity of the structural relationship;
The common expression group included in common in each sentence of the similar sentence pair is in a similar relative positional relationship in the sentence structure of each sentence, and the difference expression group that is different between the sentences of the similar sentence pair Based on the fact that the common expression group is in a similar relative positional relationship in the sentence structure of each sentence, the synonym expression extracting process for extracting the synonym expression from the similar sentence pair is executed,
In the synonym expression extraction process, a process of extracting a synonym expression based on the calculated similarity
Synonymous expression extraction program to be executed. A synonymous expression extraction program for extracting synonymous expressions from a pair of similar sentences in which similar sentences are paired,
On the computer,
A sentence structure specifying process for specifying a sentence structure of the pair of similar sentences;
Based on the specified sentence structure, a common expression that is an expression that is commonly included in each sentence of the similar sentence pair and a difference expression that is an expression that is included only in each sentence of the similar sentence pair are extracted. Expression extraction processing,
Based on the extracted common expression and difference expression, a positional relationship specifying process for specifying a relative positional relationship between the common expression and the difference expression in the sentence structure of each sentence of the similar sentence pair;
Based on the identified relative positional relationship, a similarity calculation process for calculating the similarity of the difference expression in the similar sentence pair from the identity of the structural relationship;
The common expression group included in common in each sentence of the similar sentence pair is in a similar relative positional relationship in the sentence structure of each sentence, and the difference expression group that is different between the sentences of the similar sentence pair Based on the fact that the common expression group is in a similar relative positional relationship in the sentence structure of each sentence, the synonym expression extracting process for extracting the synonym expression from the similar sentence pair is executed,
In the synonym expression extraction process, a process of extracting a synonym expression based on the calculated similarity
Synonymous expression extraction program to be executed. On the computer,
Execute a sentence conversion process that converts sentences included in a pair of similar sentences using a predetermined sentence conversion rule so that the structure of similar sentences has a similar structure.
The synonymous expression extraction program according to any one of claims 12 to 14 . On the computer,
In the synonym expression extraction process, according to a predetermined similarity criterion for extracting the synonym expression, a process of extracting the synonym expression from the pair of similar sentences is executed,
Based on correct / incorrect information indicating the correctness of the extraction result of the extracted synonym expression and the similarity of the synonym expression, a similarity criterion correction process for correcting the similarity criterion for extracting the synonym expression is executed.
The program for synonymous expression extraction according to any one of claims 12 to 15 .

Images