JP2005063332A - Information system coordination device, and coordination method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To support data integration by efficiently detecting a pair of similar information elements among different information systems. <P>SOLUTION: The information system coordination device 1 is provided with at least a characteristic analysis means 2 for analyzing statistical characteristics of data of respective information elements belonging to each information system, and an element pair detection means 3 for providing a common space for comparing a plurality of information systems on the basis of an analysis result, and detecting elements with similar statistical characteristics of data from information elements belonging to different information systems in the common space as an element pair. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to an information system association method by a computer, and as a specific field, firstly, it relates to the use / management of a text data classification system or a classification system of an arbitrary object associated with metadata. There is a field.

  Here, the text data refers to a plain text, a general document such as a word processor, a document such as a Web page, an e-mail, or a piece of character string that can be meaningful even for fragmented information.

  Metadata refers to any object that can be classified (both concrete and abstract), such as goods, tools, machines, paper or electronic media books / documents, people, organizations, etc. Original objective features, characteristics, and artificially assigned data (product prices, document transmission dates, impressions and comments about books, etc.) are structured and organized by information type. Data. As a metadata data format, individual features and the like are expressed as attribute name-attribute value pairs, and each object is expressed as an attribute data group composed of attribute name-attribute value pairs, or in XML format. Express complex attribute-attribute value relationships and meta attributes (attribute attributes) in a tag structure that matches the nested structure between attributes using RDF or RDF (Resource Description Framework) format There are methods.

  Second, there is a field related to the use and management of tag systems (hierarchical structure) of structured documents with tags such as XML, SGML, RDF, HTML, etc. Third, in the table in the relational database (RDB) system There are fields related to the use and management of the field group system, and fields related to the use and management of the object (class) attribute group system in the object-oriented database (ODB).

  The present invention provides support for data integration in such various fields, for example, data integration between different departments within a company, and data integration at the time of corporate merger.

  For example, as a classification system of books, DDC (Dewey Decimal Classification), UDC (International Decimal Classification), CC (Colon Classification), LCC (United States Library of Congress Classification), NDC (Japan Decimal Classification) ), NDLC (National Diet Library Classification) and many other classification methods that are well known at the international and national levels. In general, classification categories created based on different classification criteria are not compatible with each other due to a mismatch in category names, a difference in granularity, a difference in hierarchical structure, and the like. Accordingly, there is no compatibility of classification labels (category names) between information classified by different classification methods.

  In recent globalization and multi-vendor trends, the importance of information sharing and interoperability between multiple information systems, and the importance of incorporating and using information groups that share the same topics from different information systems It is increasing. For this purpose, it is necessary to associate classification labels (categories). For example, the correspondence between 100 classification categories is on the order of 10000, and is proportional to the square of the scale. The larger the scale, the greater the man-hours required for associating categories between systems. In such a case, support by machine processing is inevitable.

  In addition, even in specialized fields that partially overlap with the subject areas of book classification, specific classifications are required for specific academic fields and industrial fields, and if the industry, research institution, or researcher is different in the same field, the granularity of classification Are often different, and the classification is often different at a fine level, which is an obstacle to category association.

  Also, in the field of e-commerce, which has been in the spotlight in recent years, the product classification system differs depending on the industry or individual company level, which is an obstacle to complete automation of electronic transactions. In particular, e-commerce and web services are increasingly using tagged documents such as XML as product descriptions and transaction records for compatibility, but tag systems are being used by companies and groups of companies. (DTD, document type definition) are often different, and include incompatibility with classification system incompatibility.

  The same applies to the database field. In the case of a relational database, when data is shared or integrated between different existing databases, the correspondence between the systems of a plurality of table-field groups (for example, the name field of the address book table of the personnel database = the general affairs DB) To find out the relationship of employee fields in the employee stock ownership table. In the case of an object-oriented database, the association between object (class) -attribute group systems becomes a problem.

  In summary, when an information system (classification system, tag system, RDB table-field system, ODB class-attribute system, etc.) and another information system are integrated or interoperated, in many cases, Since incompatibility of information classes (classification category, tag, RDB field, ODB attribute, etc.) occurs and work man-hours that cannot be handled manually occur, it is necessary to associate different information systems by machine processing There is sex.

The following documents are available as conventional techniques for category matching methods between different classification systems.
Japanese Patent Application Laid-Open No. 10-116290 “Document Classification Management Method and Document Search Method” Japanese Patent Laid-Open No. 2001-184358, “Information Retrieval Device by Category Factor, Information Retrieval Method, and Program Recording Medium Therefor” JP 2000-250919 A "Document Processing Apparatus and Program Storage Medium Therefor" Ichise et al .: Learning rules for adjusting hierarchical knowledge, Journal of the Japanese Society for Artificial Intelligence, Vol. 17, No. 3, F. PP. 230-238 (2002)

  The technique of Patent Document 1 is based on a vector space method in units of categories. The target information type is a tagged document, and a document parameter vector including an attribute name-attribute value pair specified by a tag in the document is generated. Next, the document parameter vectors are grouped for each document class (the classification category in the present invention, or simply agreed with the category), and the centroid of those vectors is converted into the document class parameter vector (the category feature vector in the present invention). As an agreement), the similarity of the category feature vectors in the two classification systems is compared to find the correspondence between the classification categories.

  The execution order is sequential (category number order), and processing along the category tree structure is not performed. The correspondence between categories is basically a one-to-one correspondence. Since only the end category is targeted, when a one-to-many correspondence is found, the correspondence is made so that the upper and lower relationships exist between the categories of the two systems. However, one-to-many extraction is affected by the execution order and similarity error. Also, the consistency of the category correspondence as the entire classification hierarchy is not evaluated.

  Patent Document 2 discloses a technique for performing a search at a category data level instead of a character string / word level search such as a full-text search in order to absorb a difference in vocabulary between different information sources.

By extracting a topic field (= category level) from the target document base using the technique of Patent Document 3, the category hierarchy is subdivided from one hierarchy to two. The category in the second layer is a topic area, and this topic area is associated with different information sources. Mapping the degree of similarity calculated by the inner product in vector space, the category C _B category C _A and highest similarity scheme B of scheme A the corresponding topic areas (categories). The inner product calculation of the vector itself is not new, but since the category correspondence is calculated from both systems, not only the one-to-one correspondence but also the one-to-n correspondence can be found.

The technique of Non-Patent Document 1 is a method for taking an instance (document, Web page, etc.) of a knowledge source that has been classified as a concept system as an instance of another different concept system. (Here, “concept” is the same as “classification category” or simply “category” in the present invention.)
The common part of web pages already classified by two different classification systems is used as teacher information, and the similarity of categories between the two classification systems can be determined by using the consistency test (κ statistic). Have found. The structure of the classification hierarchy is a tree structure, not a lattice structure.

  It is a top-down recursive algorithm that examines the correspondence of only adjacent hierarchical relationships. Accordingly, it is not possible to find a correspondence relationship between orthogonal classification criteria and apparently distant categories. This also does not evaluate the consistency of the category correspondence as the entire classification hierarchy.

  In the prior art described above, in the category pair (category pair associated between classification systems) extraction method, similarity in vector space and consistency of data shared between two systems are tested. A single method is used. However, the former cannot handle the upper and lower relationships in the hierarchy, and the latter cannot handle the consistency of the corresponding relationship in the entire hierarchy, although it deals with the upper-lower relationship in one hierarchy. The contents (attribute information, appearance word characteristics, etc.) cannot be handled. Thus, both have advantages and disadvantages, and it is necessary to judge integrated consistency. The conventional method has not considered such a situation.

  The first problem of the present invention is to provide a common space for comparing the similarity of feature vectors in two classification systems, for example, and to add similarity of names by comparing similarities in the spaces. Further, the similarity of the information element pair, for example, the category pair detected by using the integrated similarity is further increased.

The second problem of the present invention is to evaluate the structural consistency indicating whether the positions of the elements constituting the detected element pair in the information system are consistent with each other in a plurality of information systems. It is possible to detect a highly consistent element pair set, for example, a category pair set.
That is, an object of the present invention is to efficiently support data integration by detecting pairs of similar information elements between different information systems.

  FIG. 1 is a block diagram showing the principle configuration of an information system associating apparatus 1 according to the present invention. In FIG. 1, a feature analysis unit 2 analyzes statistical characteristics of data of individual information elements belonging to each information system based on sample data for data of information elements belonging to a plurality of information systems. The detection means 3 provides a common space for comparing a plurality of information systems based on the analysis result, and statistical characteristics of element data among information elements belonging to different information systems on the common space. Are detected as element pairs.

  The information system associating position 1 further includes name similarity detection means 4 for detecting the similarity of element names between elements belonging to different information systems, and the element pair detection means 3 is similar to the statistical feature described above. It is also possible to detect an element pair having a high integrated similarity by integrating the name similarity.

  Furthermore, the information system associating apparatus 1 is a structural match that indicates whether or not the positions of the elements constituting the element pair detected by the element pair detection means 3 are consistent with each other among the plurality of information systems. Consistency evaluation means 5 for evaluating the property can also be provided.

  In the embodiment of the invention, the elements constituting the detected element pair and the elements constituting the other detected element pair among the plurality of information systems in which the consistency evaluation means 5 shows the directed graph relationship It is also possible to evaluate the consistency of the hierarchical relations including the upper-lower relations and / or the distance between elements in the information system as a structural consistency. Among them, the consistency of the neighborhood relationship including the distance between the element constituting the detected element pair and the element constituting the other detected element pair can be evaluated as the structural consistency.

  In the embodiment, the information system associating apparatus 1 further includes an optimum element pair output unit that outputs a set of request pairs having a high structural consistency as an optimum element pair set among a plurality of information systems. In addition, among the element pairs detected by the element pair detecting means 3, the element pair display means for displaying the element pairs having the highest structural consistency up to the element pair having the highest structural consistency. The element correspondence data storage means for storing the correspondence between the information elements in the plurality of information systems and the data corresponding to the elements, and the stored contents and the structural consistency of the element correspondence data storage means. It is also possible to further comprise data search means for searching for data in the same field of different information sources or data corresponding to a logical operation of the data using high element pair data.

  Further, in the embodiment, the element pair detection means 3 detects the element pair that matches the teacher data by using the teacher data of the element pair designated from the outside among the elements belonging to a plurality of information systems. You can also.

  Next, in the information system association method of the present invention, based on sample data for information element data belonging to a plurality of information systems, statistical characteristics of data of individual information elements belonging to each information system are analyzed, Based on the analysis results, a common space for comparing multiple information systems is provided, and elements with similar statistical characteristics of element data are paired between information elements belonging to different information systems on the common space. The method of detecting as is used. Further, a program for causing a computer to execute a procedure corresponding to this method and a computer-readable portable storage medium storing the program are used.

  In a field related to the classification system of the information system, an optimum set of category pairs can be obtained by reflecting the overall hierarchical structure of the classification system in the association of classification categories. This opens the way to automating category association work between different classification systems, which conventionally relies on manual work. In addition, more appropriate category pair candidates that reflect integrated similarity criteria from a plurality of viewpoints can be generated in the generation of category pair candidates.

In a field related to a structured document with a tag such as XML, an optimum set of tag pairs can be obtained by reflecting the overall hierarchical structure of the tag system in the tag association in the tag system. Further, in the generation of tag pair candidates, more appropriate tag pair candidates reflecting the integrated similarity criterion from a plurality of viewpoints can be generated.
In a field related to a database system, more appropriate field pair candidates that reflect integrated similarity criteria from a plurality of viewpoints can be generated at the time of field correspondence in a database table.

  As described above, according to the present invention, it is possible to efficiently integrate data of different information systems. For example, in the case of mergers, acquisitions, and alliances between companies, associations between large-scale databases between different departments in the company, and integration of classification systems, it is possible to realize significant cost reduction and time reduction.

  Hereinafter, embodiments of the present invention will be described in three examples corresponding to specific examples of information elements in the information system. In the first embodiment, the information system is an information classification system, and the information element is a category as a classification label.

  FIG. 2 is an explanatory diagram of category association between different classification systems. 2 are assumed to be classification systems in the same field or similar fields. Even in the same field, they are often created with different classification criteria. Each classification hierarchy is represented by a tree structure or a lattice structure. Each node in the hierarchical structure indicates one classification category in the classification system. Since the classification systems A and B are information systems in the same or similar fields, it is assumed that the categories in each classification system include substantially synonymous or similar categories. For example, the dotted arrow in FIG. 2 represents the correspondence relationship between the category A2 in the classification system A and the category B1 in the classification system B that are the same or similar. Similarly, the category pairs of A5 and B3, or A6 and B5, and A10 and B10 in the systems A and B are the same or similar categories.

  The same or similar category is easy if it can be judged from the category name alone, but generally the same words, synonyms, and synonyms are not always used, so they try to find these relationships automatically or semi-automatically. This is the object of the present invention.

FIG. 3 is a block diagram showing the configuration of the information system associating device according to the first embodiment. The control unit C10 controls the overall processing flow.
The category information storage unit (I _A and _{I B) 11 a, 11 b} , the information belonging to each category of each classification system A and B (text data and attribute names - such as AVP) is stored.

In the information hierarchy relationship storage units (H _A and H _B ) 12 _a and 12 _b , data on the upper-lower relationship of each category in the classification hierarchy of the classification systems A and B is stored.
Category feature processing unit (CC, category character squirrel Kinetics) At 13, the information belonging category information storage unit from the (I _A) 11 _a for each category, and information layer relation storage (H _A) classified 12 _a The data of the upper-lower relationship of each category in the system A is received, the feature word for each category reflecting the hierarchical structure is extracted, the category feature vector is created, and stored in the category feature vector storage unit (V _A ) 14 _a . . Similarly, the information belonging to each category from the category-specific information storage unit (I _B ) 11 _b and the upper-lower relationship of each category in the classification system B from the information hierarchy relation storage unit (H _B ) 12 _b Data is received, a category feature vector is created, and stored in the category feature vector storage unit (V _B ) 14 _b .

The category feature vector storage unit (V _A ) 14 _a and the category feature vector storage unit (V _B ) 14 _b store feature vectors generated based on text data or feature amounts extracted from metadata.

The category pair storage unit (CP, category pair) 15 stores a corresponding category pair between the classification system A and the classification system B.
The vector similarity processing unit (VS, vector similarity) 16 reads the category feature vectors from the category feature vector storage unit (V _A ) 14 _a and the category feature vector storage unit (V _B ) 14 _b , and classifies them with the classification system A. A corresponding category pair between the systems B is found and stored in the category pair storage unit CP15.

The category name similarity processing unit (LS, label similarity) 17 reads the names of individual categories from the category-specific information storage units (I _A ) 11 _a and (I _B ) 11 _b , and calculates the similarity between the category names. Calculate and store in the category pair storage unit CP15.

  The hierarchical relationship consistency processing unit HC (hierarchy consistency) 18 determines whether the categories of the category pairs stored in the category pair storage unit CP15 are consistent with each other in the hierarchical relationship between the two systems A and B. , Which is detected as hierarchical consistency.

In FIG. 3, the category feature processing unit CC13 and the vector similarity processing unit VS16 correspond to the feature analysis unit and the element pair detection unit in claim 1 of the present invention.
Further, the category name similarity processing unit LS17 corresponds to the name similarity detecting means in claim 2, and the hierarchical relationship matching processing unit HC18 is a consistency that evaluates the consistency of the hierarchical relationship as structural consistency in claim 4. Corresponds to the evaluation means.

  FIG. 4 is an overall process flowchart of category matching in the first embodiment. In the first embodiment, for example, category matching is performed using system elements of category system A and category system B having a hierarchical structure, that is, sample data of node data.

  First, in step S1, morphological analysis and hierarchical feature word extraction processing are performed using the sample data of the two systems A and B. This processing is performed using a known technique such as the above-mentioned document.

  Subsequently, in step S2, similarity matching in a multidimensional space, that is, similarity matching in a vector space, and similarity matching based on category names are performed. For example, the similarity obtained in the multidimensional space and the similarity obtained by name similarity matching are integrated, and the system B side category having a high similarity with the system A side category is combined by the integrated similarity. Are output as category pair candidates in step S3.

  Subsequently, in step S4, the structural consistency of the category pair (candidate), that is, the process of evaluating whether the positions occupied by the two categories constituting the category pair in the category system to which each category belongs is mutually matched. As a result of the processing, category pairs having high structural consistency are output as optimum category pair candidates in step S5.

  As described with reference to FIG. 4, in the present invention, sample data for each information system is necessary. In the similarity matching in the multidimensional space in step S2, the position of the category feature vector in the space is determined based on the sample data. If there is no sample data, a distribution cannot be created, and the position of the category feature vector in the space cannot be determined. Similar matching based on category names is possible even if sample data does not exist, but since the amount of information is small with only category names, matching accuracy cannot be increased.

  As a matter of course, it is better that the amount of sample data is large, but empirically, dozens of documents are required for each category. For the end node of the graph, for example, a sample of about 10 pages in a Web page. It is desirable to have data. By using sufficient sample data, the characteristics of the information system can be clarified, and the similarity of category pairs in similarity matching is improved.

  FIG. 5 is an explanatory diagram of comparison of category feature vectors between different systems. The calculation of the similarity of feature vectors used for matching the classification system of two information sources will be described. As a means for finding a category correspondence candidate between the classification system A and the classification system B, it is necessary to generate a category feature vector in the vector space. As the feature amount necessary for the category feature vector, the feature amount of the sample data for each category on the classification system is used.

  As described above, there are two types of feature amounts of sample data, and category feature vectors are created according to the types. First, when the sample data is metadata, each attribute in the metadata is a coordinate axis, and the value of the attribute is a coordinate value. This coordinate axis-coordinate value pair relationship is a vector element.

  In the case of text data, by using the technique of the following Patent Document 4 from the target text data, feature words for each category are extracted from the classification systems A and B, and the degree of association between the category and the feature word is obtained. The relationship between the coordinate axis and the coordinate value pair can be used as a vector element by associating the resulting feature word with the coordinate axis and the degree of association with the coordinate value.

Japanese Patent Application No. 2002-185173 “Feature Word Extraction System” Thereby, each category in each classification system is associated with a vector in a vector space corresponding to each classification system. In FIG. 5, a category on the classification system A is represented as a white circle of the category distribution of A on the vector space, and a category on the classification system B is represented as a black circle of the category distribution of B on the vector space.

  As it is, there is a difference between the m-dimensional coordinate axis (corresponding to the feature word) in the vector space V (A) and the n-dimensional coordinate axis in V (B), so the part that uses only the common coordinate axis part It is necessary to compare on space. For this purpose, a common part (product set) of a set of feature words in the classification system A and a set of feature words in the classification system B is obtained, and a coordinate space corresponding to the feature word set is adopted to obtain a vector space. V (A∩B) should be constructed. If vectors corresponding to the categories of the classification system A and the classification system B are arranged in this vector space, it is possible to compare the similarity of category feature vectors of different classification systems.

  Criteria for comparing the similarity of two vectors in space include cosine measure, Euclidean distance, Hamming distance, and the like. In addition, since selection of presence / absence of vector normalization (absolute value = 1) is also conceivable, these comparison conditions may be selected in accordance with the nature of the target data and the purpose of use. When the angle formed by the two vectors is α, the value of cos α calculated using the inner product (scalar product) and the absolute value of each vector is a cosine measure, and the angle α is called an angular distance.

  Regarding similarity evaluation, a smaller distance is more similar, and a cosine scale is larger and closer to 1, the two vectors are similar. Therefore, the determination may be made according to a comparison criterion. The similarity of the category feature vector is used in combination with the similarity based on other criteria in later processing. In the present invention, in order to make it easy to find similar categories among many category groups, vectors are similar so that they can contribute to the sum and product of operations often used for synthesis with other reference values. The more it is done, the larger the “category similarity by vector” (positive). Therefore, when the concept of distance is used for comparison of vectors, the category similarity by vector may be determined by “take reciprocal” or “multiply −1” as necessary. In addition, a conversion (linear conversion or the like) that adjusts the range of the category similarity by vector may be performed. Thereby, the similarity of the category feature vector can be calculated.

  Finally, it is checked whether or not the similar category pair candidate satisfies the candidate condition. For example, a threshold value regarding the similarity in the vector space is checked. For example, in the similarity measure such as cosine measure, the lower limit is defined as follows.

Sim _VECT ≧ _cosα (1)
In addition, the upper limit is defined as follows in distance scales such as Euclidean distance and angular distance.

Sim _VECT ≦ α _VECT (2)
FIG. 6 is a detailed process flowchart of the similar category detection process in the vector space described with reference to FIG. 5, that is, the similar matching in the multidimensional space in step S2 of FIG. When the processing is started in the figure, first, in steps S11 to S16 and steps S17 to S22, the category distributions in the respective vector spaces are respectively targeted for the two category systems A and B, that is, the right side and the left side in FIG. The category distribution is obtained.

  In step S11, the classification system A is an analysis target, and in step S12, it is determined whether the category data of the system, that is, the category data of the node is text data or attribute data as metadata. If it is text data, the hierarchical feature word is extracted in step S13, the category feature vector is calculated in step S14, and if it is metadata attribute data, the attribute feature in step S15. After the vectors are calculated, a category distribution is formed on the vector space V (A) for the system A in step S16.

  Similar processing is performed for the classification system B in steps S17 to S22, and after the category distribution is formed on the vector space V (B) corresponding to the system B, the two systems A and B are processed in step S23. It is determined whether or not data preparation is completed, that is, whether or not a category distribution is obtained for each of the two category systems A and B, and if not, until two category distributions are obtained. The process of step S23 is repeated.

  If it is determined in step S23 that category distribution data for the two systems has been obtained, common feature words for the respective vector spaces V (A) and V (B) are obtained in step S24. A comparison space common to the systems is formed. In step S26, similar category pairs are detected and compared between the two systems A and B, and the closest category pair in the center of FIG. In step S27, it is checked whether the similar category pair satisfies the candidate condition, for example, whether it satisfies the expression (1) or (2), and the process ends.

  As described above, in this embodiment, a comparison space common to vector spaces corresponding to the two systems is formed, and similar category pairs are detected. That is, in this embodiment, the vector space method is applied to not only the terminal category but also the non-terminal category, and similar category pairs are extracted by reflecting all the hierarchical relationships from the terminal category to the top root category. The comparison of categories in the common vector space in which only the common coordinate axis portion is adopted between the m-dimensional coordinate space for the system A and the n-dimensional coordinate space corresponding to the system B is a major feature of the first embodiment. Compared to the case where only one vector space is used in the prior art, the accuracy of matching between two information systems is greatly improved.

  Next, similar matching based on category names in step S2 of FIG. 4 will be described. In this process, the character string level identity or similarity of the category names in the classification systems A and B is determined, and the synonym synonym dictionary is referred to.

A character string of the name of category a _i in classification system A is name (a _i ), and a character string of the name of category b _j in system B is name (b _j ). The identity of strings means name (a _i ) = name (b _j ), that is, an exact match. The similarity of character strings is determined based on the fact that one is the other partial character string or the number of common parts in the set of characters that constitute the character string.
For example, the case classification shown in FIG.

Set. In FIG. 7, the symbol “*” in the lower inclusion formula indicates an arbitrary character string. For example, when the name of a _i is “applied number” or “basic mathematics” and the name of b _j is “mathematics”, a _i is a subordinate category of b _j . The meaning of upper inclusion or intermediate inclusion is the same, for example, “math exercise” is a subcategory of mathematics.
For example, “word” in the lower partial matching word means a word already registered in the dictionary heading. This dictionary may be a synonym synonym dictionary, a morphological analysis dictionary, another electronic dictionary, or a dictionary combining these dictionaries. The fact that a _i and b _j are siblings means that the nodes corresponding to a _i and b _j share the most recent node in the hierarchical relationship of category nodes, as will be described later. The cousin relationship also means that the two nodes are not in the immediate upper level, but have a common node other than the root node in the upper level.

“Str” in the lower part match also means an arbitrary character string, but this character string is common to the names of a _i and b _j , and the arbitrary character string indicated by the symbol * is two names. Are different.

The similarity value at the character string level is determined with reference to the category relationship expected for the type of similarity. For example, it is determined as follows.
γ _eq = 0.9, γ _li = 0.8, γ _ui = 0.7, γ _mi = 0.4, γ _lpw = 0.6, γ _lps = 0.5, γ _pw = 0.3, γ _oo = 0.2, γ _o = 0.1
The number of characters in str, the composition ratio of common characters, the appearance order matching rate of common characters, or the like may be set as a variable numerical value.

When a synonym synonym dictionary can be used, the dictionary is preferentially used rather than the similarity calculation at the character string level. FIG. 8 shows a configuration of the synonym synonym dictionary.
The synonym synonym dictionary includes a character string as `` representative word '', a character string as `` synonym synonym '', a value of `` similarity '' (0 ≦ x ≦ 1), `` registration date '', presence of `` AUTHORIZED '', `` field It consists of “information”, “polysemy” and the like. Of these, the synonym and synonym of the representative word and the items of similarity are essential (indicated by *). A representative word is one of the elements of a set of synonyms, and a word that represents the set of synonyms is selected. Write the synonym synonym item in the synonym synonym field, excluding the representative word. AUTHORIZED is set to 1 when the dictionary administrator's organization or group has an agreement, otherwise it is set to 0. In other words, the AUTHORIZED value of the target data is 0 at the stage of personal judgment at the person-in-charge level. If an agreement is reached, the registration date value should be changed to the agreed date. The field information may be a specialized field name such as politics, economy, IT, medicine, and general daily life, or a category name of an appropriate hierarchical classification system. In the subject field written in the field information, the synonym is written when the synonym has a plurality of meanings, and is set to 0 in other cases.

In addition, the similarity value is determined under appropriate constraints based on the judgment of the operator or the dictionary management organization. For example, it is determined as follows.
Synonym: 0.9 ≦ x ≦ 1.0 (3)
Similarity: α _NAME ≦ x ≦ 0.9 (4)
Here, α _NAME (≧ 0) is a threshold value of name similarity, and in order to become a category pair candidate based on name similarity, the following conditions must be satisfied.

Sim _NAME ≧ α _NAME (5)
The determination of the similarity of category names using the synonym synonym dictionary is performed as follows. For this purpose, the determination method of synonymity and similarity shown in FIG. 9 is used. The two words (character strings) to be investigated are word ₁ and word ₂ . When word ₁ and word ₂ satisfy any of the following conditions, it is determined that there is synonymity or similarity. This is referred to as lexical similarity Sim _DIC of word ₁ and _{word 2 (word 1, word 2} ).
・ If _one of word ₁ and word ₂ is a representative word and the other is a synonym for that representative word (similarity is the similarity of the synonym)
・ When both word ₁ and word ₂ are synonymous synonyms for the same representative word (similarity is the smaller of those synonyms)
When such a synonym synonym dictionary is available, the category name similarity determination process shown in the flowchart of FIG. 10 is performed. In FIG. 10, the similarity between the name of the category a _i in the system A and the name of the category b _j in the system B is determined.

  First, in step S31, it is determined whether or not the names of the two categories are registered in the synonym synonym dictionary described with reference to FIG. If the determination result of the sex is the similarity of the category name and is not registered in the dictionary, the determination result of the similarity of the category name based on the character string similarity is set as the similarity of the category name in step S33. In step S34, as a check of candidate conditions for similar category pairs, it is determined whether or not equation (5) is satisfied, and the process is terminated.

  Note that the threshold value used in equation (5) is determined by experiment, for example. As the method, for example, a set of category pairs with high structural consistency (correct answer) described later is prepared, and a category pair set with a high degree of similarity obtained by changing the threshold value is more accurate. It is conceivable to adopt a value that gives

Next, the integrated similarity obtained by integrating the similarity of the category pair on the vector space described with reference to FIGS. 5 and 6 and the similarity of the names described with reference to FIGS.
The similarity between the systems A and B is the similarity between the categories a _k and b _{l in} the vector space defined by the equations (1) and (2)

Or the similarity between the names of the categories a _k and b _l obtained in step S22 or S23 of FIG.

Therefore, a category pair similar between systems A and B can be obtained based on this. In addition, by setting a similarity obtained by integrating these two different criteria, it is possible to set the similarity between categories to be higher when both the statistical characteristics between categories and the similarity of names are high. For example, using the integrated similarity defined by the following equation (6), similar category pair candidates can be found from an integrated viewpoint.

： Similarity between categories a _k and b _l in vector space

: Similarities between names of categories a _k and b _l

: Vector similarity weight (> 0),

: Weight of name similarity (> 0)
Subsequently, the evaluation of the structural consistency of the category pair and the category pair set detected by the similarity calculation will be described. In the calculation of the similarity by the vector space method, the hierarchical relationship (upper-lower relationship) between categories in the same classification system is reflected in the calculation by using the technique of Patent Document 4 described above. However, regarding the obtained category pair, it is impossible to know information on the consistency between the category position in the hierarchical relationship in one system A and the consistency of the category position in the hierarchical relationship in the system B on the other side of the category pair. In order to obtain an optimal solution, it is necessary to find a set of category pairs that have a corresponding relationship in which the hierarchical relationship as a whole of the category pair is best applied.

  The relationship between individual categories in the classification system includes a hierarchical relationship (directed graph relationship) and a neighborhood relationship (undirected graph relationship). Hierarchical relationships include upper-lower relationships and overall-partial relationships. Hierarchical relationships are represented by directed graphs (individual nodes connected by links with arrows), and neighborhood relationships are represented by undirected graphs (individual nodes connected by links without arrows).

  Therefore, for the consistency of category relationships, the consistency of neighborhood relationships should be considered as necessary in addition to the consistency of hierarchical relationships. The concept of consistency calculation will be described below for each case. In addition, the consistency of the hierarchical relationship and the consistency of the neighborhood relationship are collectively referred to as structural consistency. First, the consistency of the hierarchical relationship will be described with reference to FIG. A solid line and a dotted line in FIG. 11 are category pair candidates, and are obtained by the above-described method. In this section, we will build a mechanism to comprehensively judge whether a given category pair as a whole fits well in the upper-lower relationship of the hierarchy of the two classification systems, or whether a twisting phenomenon occurs. .

  If there are classification systems A and B, and category A3 in system A and category B6 in system B are listed as similar category pair candidates, the correspondence between these two categories is A mechanism for determining whether or not the category pair A3-B6 is in the correct correspondence by evaluating whether or not there is consistency compared with the inside position (is better fit) will be explained. Here, the category pair A3 to B6 to be evaluated is referred to as a reference category pair, and the categories A3 and B6 constituting the reference category pair are referred to as reference categories. Note that, as will be described later, in the plurality of obtained category pairs, that is, the category set, each category pair is sequentially set as a reference category pair, and the consistency is evaluated.

  For example, for category pair (1) in FIG. 11, category A1 is one level higher than standard category A3, and category B2 on system B paired with A1 is exactly the same as standard category B6. There is a one level higher relationship. Therefore, since categories A1 and B2 related to category pair A1-B2 are both in the same upper hierarchy with respect to their respective reference categories, as far as these two category pairs are concerned, each hierarchical structure of systems A and B It can be seen that the consistency is very good.

  In addition, regarding category pair (2) in FIG. 11, category A1 is one level higher than standard category A3, and category B9 on system B paired with A1 is opposite to standard category B6. Is one level lower. Therefore, since the categories A1 and B9 related to the category pair A1-B9 are in the opposite hierarchical relationship with respect to the reference category, the hierarchical structures and twists of the systems A and B occur as far as these two category pairs are concerned. This shows that the consistency is poor.

  Next, with respect to the category pair (3) in FIG. 11, there is a category A1 one level higher than the reference category A3, and the category B7 on the system B paired with A1 is different from the reference category B6. There is a relationship. Here, the cousin relationship refers to the case where two categories share the same upper category (excluding the root category). When two categories share a direct upper category, it is called a sibling relationship, but in the present invention, it is unified by the term cousin relationship as a broader concept. Therefore, since the categories A1 and B7 related to the category pair A1-B7 are in a one-sided relationship, one-level relationship, one-sided cousin relationship with respect to the reference category, as far as these two category pairs are concerned, each hierarchy of systems A and B At first glance, it is not clear whether the structure and consistency are good or bad. In such a case, it is necessary to evaluate the hierarchical consistency of the category pair A1-B7 based on the hierarchical relationship and link distance. Here, the link distance is the number of links that go through to reach the reference category from the category, and if there are multiple routes from the category to the reference category, the links that go through them The number of links of the route with the smallest number is the distance.

  In this way, since the type of hierarchical consistency of the category pair is different from the type of the category relationship with respect to the reference category, these are then organized together. FIG. 12 is a diagram for explaining the evaluation of hierarchical consistency of two pairs of category relationships between two classification systems. Here, the category relationship with respect to the reference category is divided into four types, upper, lower, cousin and irrelevant. However, irrelevant here means that the two categories do not share a category other than the root category as an upper category.

  In the figure, in the case of higher order on the system A side, as described above, the category relationship matches as higher in the case of higher order on the system B side, the reverse order in the lower order on the system B side, and the system B side. In the case of a cousin relationship, it is irrelevant if it is irrelevant on the system B side.

  Next, the lower case on the system A side is considered in the same way as the upper case. In other words, the category relationship is reverse in the case of higher order on the system B side, the lower order on the system B side is the same as the lower order, the case of cousin relation on the system B side, and the case of irrelevance on the system B side is irrelevant. .

  Furthermore, in the case of a cousin on the system A side, if it is higher or lower on the system B side, the cousin relation on the system B side is the same as the cousin relation, and if it is irrelevant on the system B side, it is irrelevant. Become.

When the system A is irrelevant, the result is irrelevant regardless of the system B side.
Consistency when focusing only on two pairs of categories between two classification systems is evaluated as a hierarchical fitness. In calculating the hierarchical fitness of the actual category pair, the fitness is determined in consideration of appropriate weighting for matching, reverse order, irrelevant, etc. in FIG.

For example, in order to prioritize the category pair having the same hierarchical relationship and avoid the reverse order relationship as much as possible, the hierarchical fitness of a _k −b _l with respect to the reference category pair a _i −b _j

Can be set as follows.
Match (Top): 1.0
Match (subordinate): 1.0
Match (cousin): 0.4
Reverse order: −1.0
Other: 0.1
Irrelevant: 0.0
Alternatively, it can be increased or decreased according to the link distance as follows. Where λ (> 0) is the link weight, l _a and l _B are the link distances to the reference category,

And
Match (top)

Upper link weight: λ _sup > 0
Match (subordinate)

Lower link weight: λ _sub > 0
Match (cousin)

Cousin relation weight

Reverse order:

Reverse order weight: λ _rev > 0
Other

Other relationship weights

Irrelevant: 0.0
As described above, the hierarchical suitability can be appropriately determined. Various weights used here can also be determined experimentally.

Hierarchical consistency of category pair set Ω

First, the hierarchical consistency of the category pair a _k- b _l with respect to the reference category pair a _i- b _j

Is obtained by the following.

: Hierarchical fitness of a _k- b _l for category pair a _i− b _j Next, hierarchical consistency of category pair a _i− b _j

Is obtained as follows.

Ω: set of category pairs,
| Ω |: Size of category pair set (number of elements in the set)
Finally, hierarchical consistency can be determined for the entire category pair as follows.

This hierarchical consistency

The set of category pairs that maximizes is the optimal solution from the viewpoint of hierarchical relationships.
Next, the consistency of the category pair and the neighborhood relationship of the category pair set will be described. FIG. 13 is an explanatory diagram of the consistency of neighborhood relations between category pairs. The solid and dotted lines in FIG. 13 are category pair candidates, and are obtained by the above-described method. In this section, we construct a mechanism for comprehensively determining whether a given category pair as a whole fits well with the neighborhood relationship between the two classification systems or causes a twisting phenomenon.

  If there are classification systems A and B, and category A3 in system A and category B6 in system B are listed as similar category pair candidates, the correspondence between these two categories is the undirected graph. Explains the mechanism for determining whether the category pair A3-B6 is in the correct correspondence by evaluating whether it is consistent (or fit) compared to the position in. Here, the category pair A3 to B6 to be evaluated is referred to as a reference category pair, and the categories A3 and B6 constituting the reference category pair are referred to as reference categories.

  For example, with respect to the category pair (1) in FIG. 13, the reference category A3 and the category A1 are connected by one link, and the category B2 on the system B paired with A1 is changed to the reference category B6. Just one link away. Therefore, both categories A1 and B2 related to category pair A1-B2 have the same link distance = 1 relationship with respect to their respective reference categories. It turns out that it is very consistent with the direction graph.

  In addition, regarding the category pair (2) in FIG. 13, the reference category A3 and the category A1 are connected by one link, and the category B9 on the system B paired with A1 is changed to the reference category B6. Again, it is one link away. Therefore, both categories A1 and B9 related to category pair A1-B9 have the same link distance = 1 relationship with respect to their respective reference categories, so as far as these two category pairs are concerned, there is no difference between systems A and B. It turns out that it is very consistent with the direction graph.

  Then, with respect to the category pair (3) in FIG. 13, the reference category A3 and the category A1 are connected by one link, and the category B7 on the system B paired with A1 is changed to the reference category B6. In contrast, three links are connected. Accordingly, the categories A1 and B7 related to the category pair A1−B7 have different link distances with respect to the reference category. Therefore, these two category pairs are different from the category pairs (1) and (2) in the system A. , B are not consistent in the undirected graphs. In order to compare the degree of consistency, evaluation may be performed based on the link distance.

For example, the link distance to the reference category a _i category a _k in scheme A and l _A, the link distance to the reference category b _j of categories b _l forming the a _k and pairs in scheme B When l _B, categories versus a _The degree of neighbor relationship matching for the reference category pair of _k −b _l can be expressed as follows.

That is, the smaller the link distance difference, the larger the fitness value, the maximum value is 1, and the minimum value is 0.
In the above formula, the close distance match and the long distance match have the same evaluation, but there is also an idea that the evaluation value of the category pair whose distance to the reference category pair is close should be increased. In this case, for example, by setting the following evaluation formula, it is possible to give priority to the matching of distances in the neighborhood rather than far away.

For example, in the case of λ ₁ = 0.05 and λ ₂ = 0.1, the neighborhood relationship suitability is a value shown in FIG.
It can be seen from FIG. 14 that distance matching in the neighborhood is emphasized. In this case, when the link distance is greater than 5, the evaluation value is 0 even if the link distances match. Therefore, it is only necessary to calculate the link distance of the neighboring portion, which contributes to improvement of calculation efficiency.

Similar to Equations (7) to (9) for finding hierarchical consistency based on the degree of matching of the neighborhood relationship with respect to the reference category pair a _i -b _{j of} an arbitrary category pair a _k −b _{l in} the category pair set Ω. Neighbor relationship consistency is required.

Neighbor consistency of a set of category pairs Ω

For the reference category pair a _i −b _{j and} the neighborhood pair consistency of the category pair a _k −b _l

Is obtained by the following.

: Neighbor relations fitness categories versus a _k -b _l for categories versus a _i -b _j Next, neighbor relations integrity categories versus a _i -b _j

Is obtained as follows.

Ω: set of category pairs,
| Ω |: Size of a set of category pairs Finally, neighborhood relationship consistency can be obtained for the entire category pair as follows.

This neighborhood consistency

The set of category pairs that maximizes is the optimal solution from the viewpoint of the neighborhood relationship.
In addition, by integrating the hierarchical consistency and the neighborhood relation consistency, it is possible to obtain an optimal solution from both viewpoints. This structural integrity

Can be determined, for example, as follows.

FIG. 15 is a flowchart of processing for outputting an optimum category pair set that optimizes the structural consistency of the entire set while exchanging the category pairs in the category pair set.
When the processing is started in the figure, first, in step S41, for example, using the similarity by vector, the category having the highest similarity ranking is combined with each category, and such a category pair is closest. After the set Ω is generated as a category pair candidate and “0” is substituted for the optimum value of structural consistency CON _MAX for the set Ω, the process proceeds to step S42 for obtaining the consistency of the category pair set Ω. To do.

  Here, an example in which the hierarchical consistency described in the expressions (7) to (9) is obtained as the structural consistency will be described, but the neighborhood relation consistency described in the expressions (12) to (14) may be obtained. Of course, the structural consistency described in Equation (15) that integrates the two consistency may be obtained.

As a process for obtaining the consistency of the category pair set Ω, the category pair a _i -b _j is set as a reference category pair in step S42, the reference category pair is changed, and the processes in steps S43 to S45 are repeated.

In step S43, the process of step S44 is executed while changing the category pair other than the reference category pair a _i -b _j other than the category pair a _k -b ₁ in the category pair set Ω. In step S44, the consistency of the category pair a _k -b ₁ with respect to the reference category pair a _i -b _j , here, the hierarchical consistency given by the equation (7), is obtained. In S45, the hierarchical consistency of the reference category pair a _i -b _j , that is, the value of the equation (8) is obtained, and when the repetition of changing the reference category pair in step S42 is completed, the process proceeds to step S46. .

In step S46, the structural consistency of the entire category pair set Ω, here, the hierarchical consistency CON (Ω) given by the equation (9) is obtained, and the consistency value obtained in step S47 is the optimum consistency. It is determined whether it is greater than the value CON _MAX . Its value in step S48 is substituted for the optimum value CON _MAX if large, the set Omega is assigned to the category pair set optimal solutions Omega _MAX.

Here, since CON _MAX is set to “0” in step S41, the hierarchical consistency obtained in step S46 is set as the consistency optimum value, and the process proceeds to step S49. If, in step S47, for example, the value of the consistency optimum value stored in the memory is larger than the consistency value obtained in step S46, the process immediately proceeds to step S49.

  In step S49, the end condition is determined. As the end condition, for example, the end of a predetermined number of repetitions, the difference between the hierarchical consistency obtained in the previous step S46 and the hierarchical consistency obtained this time becomes smaller than a predetermined value, Alternatively, it can be considered that the absolute value of the increase / decrease rate of the hierarchical consistency becomes smaller than a predetermined value.

  If it is determined that the end condition is not satisfied, category pair candidate replacement processing is performed in step S50. That is, one category pair in the category pair set Ω is deleted in step S51, exchange with another category pair, addition of a category pair, and the like are performed. In step S52, the new category pair set is set as Ω. After that, the processing after step S42 is repeated.

If it is determined in step S49 that the termination condition is satisfied, the structural consistency optimum value CON _MAX and the category pair set optimum solution Ω _MAX are output in step S53, and the process is terminated.
As described above, instead of the hierarchical consistency as the structural consistency, the process of FIG. 15 can be executed using the neighbor relation consistency, and the hierarchical consistency and the neighbor relation consistency are integrated ( The processing shown in FIG. 15 can also be executed using the structural consistency given by the equation (15).

  In the process for obtaining the structural consistency of the set Ω in steps S42 to S46 after the category pair candidates are replaced in step S51 of FIG. 15, only the portion related to the replaced category pair is subject to calculation. Thus, the calculation efficiency can be improved.

  Hereinafter, claim 9 of the present invention, that is, utilization of teacher data will be described. The purpose of the present invention is to obtain a matching category pair or a similar category pair between different classification systems, but a part of the correct category pair is for some reason (for example, judgment by an expert). It may be known. In such a case, as a matter of course, a method of removing a known correct category pair from the target data and applying the similar category pair determination method only to the remaining data to find the remaining category pair is also conceivable.

  However, as another method, if the similar category pair determination method is applied to the entire data including the data of the known correct category pair to obtain the similar category pair as the overall analysis result, the known correct answer is obtained. Since the results with category pairs can be matched, if there is a difference, it becomes a clue to change the parameter or expression of the evaluation criteria or the parameter or expression of the synthesis of individual similarity so that the difference is reduced .

  If a difference between the teacher information and the actual result is required, there is a technique called machine learning in the field of artificial intelligence that finally obtains the optimum result by repeatedly changing the evaluation criteria automatically. By selecting and applying an appropriate machine learning method, a more appropriate category pair can be obtained as a result. As a result, when some of the correct answers are known in advance, a better result can be obtained than when there is no clue to the correct answer.

Claim 6 of the present invention, that is, the output of the optimum category pair set will be described.
In FIG. 15 described above, after obtaining the optimum category pair set, all of the optimum category pair sets are regarded as correct (can be said to be the same or similar category between systems A and B), and the result is output. To do. As an output destination, a display device, a file on a storage medium, a structure that can be transferred between programs, or the like can be specified. Under this condition, various sets of effects can be obtained by automatically using the set of output optimum category pairs in cooperation with other programs and network communication sockets.
For example, if there is a classification system A and it is linked with a program for integrated management of classified information, among categories in different classification systems B, a category associated with a specific category in system A Automatically copies information such as documents or web pages belonging to the category in the system B to the corresponding category in the system A, and then refers to the information in the same manner as the information originally existing in the system A. The merit that search, various analyzes, etc. can be performed arises.
As another method of realizing automation, category pair candidates obtained by similarity in vector space and similarity of category names or the two similarities are integrated without evaluating structural consistency. The category pair candidate obtained by using the integrated similarity of equation (6) is regarded as a correct answer, and the result is output to the same output destination as described above, and various effects can be obtained by linking with other programs. A method of obtaining is also conceivable.

  Claim 7 of the present invention, that is, display of category pairs with high consistency will be described. In FIG. 15 described above, not only the optimum category pair set but also the information on the category pair set having a relatively high structural consistency is stored, and the category pair set and the category pair ranking in the order of the higher structural consistency. And the ranking result is displayed on the display device.

  As information on category pairs, on the screen, ranking ranking, category names on both sides of category pairs, similarity in vector space, similarity of category names, integrated similarity, hierarchy relationship on system A side, system B side Display a hierarchical relationship, hierarchical fitness, hierarchical consistency, neighborhood relationship fitness (if any), neighborhood consistency, a list of category pair set identifiers, and the like.

  Further, as the information on the category pair set, a ranking ranking, a category pair set identifier, a structural consistency of the category pair set, a common category pair list, a non-common category pair list, and the like are displayed. Here, the common category pair is a category pair that exists in common among category pair sets having different rankings, and the non-common category pair is, for example, a category pair that exists only in one category pair set.

In addition, a diagram representing the hierarchical structure of the classification system A, a diagram representing the hierarchical structure of the classification system B, a category pair corresponding between the systems A and B, and the like are displayed.
Initially, only the category pairs belonging to the optimum category pair set are highlighted on the screen. The user can add a category pair that he / she desires to be desirable based on his / her judgment or delete a category pair which he / she does not desire while confirming such information on one screen or a plurality of screens. This function of adding / deleting category pairs can be realized by an interactive interface at both a character string level and a graphic level.

  According to the information input by the user, the contents of the set of category pairs are changed, and the display contents on the screen are changed accordingly. In addition, information regarding the contents of changes in the category pair set, the state after the change, and the editing history are also stored in the system so that they can be reused.

  As another method of realizing automation, category pair candidates obtained by similarity in vector space and similarity of category names or the two similarities are integrated without evaluating structural consistency. It is also conceivable to obtain the top n categories of category pair candidates obtained from the integrated similarity, and display and edit the results at the same output destination as described above.

Claim 8 of the present invention, that is, a data search method will be described. Here, it is assumed that a document group corresponding to a category related to a word input for search is searched.
In FIG. 15 described above, after obtaining the optimum category pair set, it is assumed that all of the optimum category pair sets are correct, and the categories among a plurality of classification systems are associated. In addition to the original classification system, a common category table reflecting the correspondence of categories is created. The items in the table include a category identifier on the system A side, a category identifier on the system B side, a hierarchical fitness, and the like.

  FIG. 16 is a block diagram showing the configuration of an information system association apparatus for executing this data search process. In the information system associating apparatus in FIG. 3, a portion irrelevant to the search process is omitted, and blocks necessary for the search process are added.

In FIG. 16, a common category table storage unit CC (common category) 20 for storing three types of tables to be described later for data search processing, and a document-category index (table) storage unit DC (document category) for category system A _A 21a, storage unit 21b for system B, word-category index (table) storage unit WC (word category) 22 and index creation unit 23 for creating these three types of tables, for example, processing search requests from users Search request processing unit 24, and a table stored in three types of tables 20, 21a, 21b, 22 corresponding to the search request, and a document group corresponding to the category related to the input word as a search result The category level search unit 25 that outputs as And a search result display unit 27 which displays the search result storage unit 26, and the search results. Note that the search request processing unit 24 uses the contents of the word-category index storage unit WC22 to determine whether or not a search using a search keyword input by the user is possible, for example, prior to the actual search. It can be referred to.

FIG. 17 shows the data structure of the common category table. When the categories in the optimal category pair set have a one-to-one correspondence (CC.1, CC.2, etc.), the category identifier (first category ID) on the system 1 side of the category pair and the system 2 side category (Second category ID ₁ ) and the structural consistency of the category pair are stored in the corresponding part of the common category table. When the category pair is one-to-many correspondence (CC.51, CC56, etc.), only one category side is a first category ID, and a plurality of category sides are a second category ID ₁ , second category ID ₂ , ... and so on. Similarly, structural consistency ₁ and structural consistency ₂ are stored.

  For a category that is not the category of the category pair in the optimal category pair set, an isolated category (C.A.2, etc.) on the system A side is replaced with the value (C .A.2 and CC.58 etc.). Other items in the same record are blank. Further, an isolated category (C.B.4, etc.) on the system B side is stored as a value of an identifier item (C.B.4, CC.97, etc.) of the category on the system B side of the common category table. This also leaves other items in the same record blank.

Next, a document-category index table is created. FIG. 18 shows the configuration of the document-category index table. For example, the correspondence between each category in the system A and documents belonging to the category is made into an index table. That is, the items of the index table include a category ID and a document-conformance list. The latter includes an identifier of the belonging document (document ID ₁ , document ID ₂ ,...) And a category conformance (conformance ₁ , conformance). Degree ₂ ... The other system B and the like are created in the same manner as the system A document-category index table.

  As for the degree of matching between a document and a category, the degree of matching between a document and a category factor is described in FIG. 5 and [0048] of Patent Document 2 described above, and the category factor may be read as a category.

  FIG. 19 shows the data structure of the word-category index table. The table items include a word ID, a word notation, an identifier (common category ID) in the common category table, and a degree of association between the word and the category. Of these, the degree of association between a word and a category is described as the degree of association between a feature word and a category in the aforementioned Patent Document 4, and the explanation can be used by replacing the feature word with a word. .

  In the word-category index table, identifiers and notations of all feature words related to each category in the common category table are stored as a word ID and a word notation, respectively. For example, these notations are stored for all feature words of each category in the system A, and these notations are also stored for all feature words of each category in the system B. As the category ID, the value of the corresponding common category ID in the common category table is stored.

Regarding the degree of association, the weight of the degree of association between each category (first category ID, second category ID list) according to the system related to the common category ID = C _k ^* on the common category table and the word, that is, the feature word The average is determined and stored as the degree of association, γ (w _i , C _k ^* ).

That is, the relevance calculation is performed as follows, for example.
(1) When one or more category pairs corresponding to the common category C _k ^* exist,

here

: A set of category pairs for the common category C _k ^* ,

: Structural consistency of category pair a _i -b _j belonging to C _k ^* (2) When there is no category pair corresponding to common category C _k ^* and one category a _i is isolated, that is, C _{If k} ^* = a _i ,

  As described above, since necessary index table information is prepared, a list of documents belonging to a category related to a search input word can be obtained by sequentially referring to the table. That is, 1) word-category index table (word → common category table identifier), 2) common category table (common category table identifier → system-specific category identifier), 3) document-category index table (system-specific category identifier → document identifier) The target document group can be specified by following the index table in this order.

  In the above description, for example, in response to a keyword input from a user, a document group corresponding to a category related to the keyword is searched. However, the search target is not limited to the document group, and various types of formats are used. Of course, it is also possible to use data, and it is also possible to retrieve data as a result of performing logical operations on data other than such documents.

  Next, a second embodiment will be described. In the second embodiment, the information system is a tag system of a structured document with a tag such as XML, SGML, HTML or the like. Tag systems A and B in FIG. 20 are tag systems in the same field or similar fields. Even in the same field, they are often designed according to different standards. Each tag hierarchy is represented by a tree structure or a lattice structure. Each node in the hierarchical structure indicates one tag in the tag system. Since the tag systems A and B are information systems in similar fields, it is assumed that tags in each tag system include substantially synonymous or similar tags. For example, the dotted arrow in FIG. 20 indicates that the tag <a1> in the tag system A and the tag <b1> in the tag system B are the same or similar tags. Similarly, the tag pairs of <a3> and <b5>, <a5> and <b6>, and <a6> and <b2> in the systems A and B are the same or similar tags.

  It is easy if the same or similar tags can be determined from the tag name alone, but generally the same words, synonyms, and synonyms are not always used, so they try to find these relationships automatically or semi-automatically. This is the purpose of the second embodiment.

  The tag system of the structured document in the second embodiment has basically the same structure as the information system as compared with the category system in the first embodiment. Accordingly, the description of FIGS. 2 to 19 as the first embodiment can be basically used as it is. By replacing the classification system in the first embodiment with the tag system and the category with the tag, for example, FIG. The information system associating apparatus and the technique including the flowchart of the entire matching process described in FIG. 4 can be used as they are in the second embodiment.

Next, a third embodiment, that is, an embodiment in which the information system is, for example, a relational database table and the information element is a field thereof will be described.
The database tables A and B in FIG. 21 are database tables in the same field or similar fields. Even in the same field, they are often designed according to different standards. Since the database tables A and B are information systems in similar fields, it is assumed that the fields in the respective database tables include substantially the same or similar fields. For example, the dotted arrow in FIG. 21 indicates that the field a1 in the database table A and the field b2 in the database table B are the same or similar. Similarly, the field pairs of a3 in system A and b3 in system B are the same or similar fields.
It is easy if the same or similar fields can be determined from the field names alone, but generally, the same words, synonyms, and synonyms are not always used, so trying to find these relationships automatically or semi-automatically. This is the purpose of the third embodiment.

The field system of the database table in the third embodiment is basically the same as the information system as compared with the category system in the first embodiment.
However, the field system of the database table in the relational database system has a flat structure without a hierarchical structure like the category system as the classification system described in FIG. Therefore, the optimum element pair utilizing the structural consistency evaluation described in FIGS. 11 to 15, that is, the optimum field pair detection technique cannot be used.

  Other techniques can be used as they are by replacing the classification system in the first embodiment with a database table and categories as fields. Although the third embodiment has been described here for the field system in the relational database, it is natural that the database may be an object-oriented database and the information system may be an object (class) attribute group system. is there.

  The information system associating device and the associating method of the present invention have been described in detail above, but this information system associating device can naturally be configured based on a general computer system. FIG. 22 is a block diagram showing the configuration of such a computer system, that is, a hardware environment.

  In FIG. 22, the computer system includes a central processing unit (CPU) 30, a read only memory (ROM) 31, a random access memory (RAM) 32, a communication interface 33, a storage device 34, an input / output device 35, and a portable storage medium reading device. 36, and a bus 37 to which all of these are connected.

  As the storage device 34, various types of storage devices such as a hard disk and a magnetic disk can be used. Such a storage device 34 or ROM 31 is shown in the flowcharts of FIGS. 4, 6, 10, and 15. And the program of claim 16 of the present invention is stored, and when such a program is executed by the CPU 30, the feature vectors in the common vector space in this embodiment are compared and named. It is possible to detect a category pair based on the similarity and to evaluate the structural consistency of the detected category pair.

  Such a program is stored in, for example, the storage device 34 from the program provider 38 side via the network 39 and the communication interface 33, or stored in a portable storage medium 40 that is commercially available and distributed. It can also be set in the reading device 36 and executed by the CPU 30. As the portable storage medium 40, various types of storage media such as a memory card, a CD-ROM, a flexible disk, an optical disk, a magneto-optical disk, and a DVD can be used, and a program stored in such a storage medium can be read. By being read by the apparatus 36, it is possible to detect an optimum category pair set having a high structural consistency in the present embodiment.

(Supplementary note 1) In an information system associating device for examining matching between systems for a plurality of information systems,
Feature analysis means for analyzing statistical characteristics of data of individual information elements belonging to each information system based on sample data corresponding to data of information elements belonging to a plurality of information systems;
Based on the analysis result, a common space for comparing a plurality of information systems is provided, and elements having similar statistical characteristics of the data of the elements among information elements belonging to different information systems on the common space An information system associating device, comprising: an element pair detecting means for detecting an element pair as an element pair.

(Supplementary note 2) The information system associating apparatus further includes name similarity detection means for detecting similarity of element names between information elements belonging to different information systems,
The information according to supplementary note 1, wherein the element pair detection means detects an element pair having a high integrated similarity by integrating the similarity of the statistical characteristics of the data of the element and the similarity of the names. System association device.

    (Supplementary Note 3) In the information system associating device, the position of the element constituting the element pair detected by the element pair detecting unit is within the system of the element constituting the other element pair detected. The information system associating device according to supplementary note 1, further comprising: a consistency evaluating unit that evaluates structural consistency indicating whether or not the position is mutually matched.

    (Additional remark 4) The system of the element which comprises the detected element pair, and the element which comprises the other detected element pair among the several information systems in which the said consistency evaluation means shows a directed graph relationship The information system associating device according to supplementary note 3, wherein the consistency of a hierarchical relation including a higher-lower relation and / or a distance between elements is evaluated as the structural consistency.

    (Additional remark 5) The said consistency evaluation means is the element which comprises the detected element pair, and the element which comprises the other detected element pair among the some information systems which show undirected graph relations 4. The information system associating device according to supplementary note 3, wherein the consistency of neighboring relations including the distance is evaluated as the structural consistency.

(Additional remark 6) In the said information system matching apparatus,
The information system association according to appendix 3, further comprising: an optimum element pair output means for outputting the set of element pairs having high structural consistency as the optimum element pair set among the plurality of information systems. apparatus.

(Supplementary note 7) In the information system associating device,
Among the element pairs detected by the element pair detecting means, the element pair having the highest structural consistency from the element pair having the highest structural consistency evaluated by the consistency evaluating means. The information system associating device according to supplementary note 3, further comprising element pair display means for displaying

(Additional remark 8) In the said information system matching apparatus,
Element correspondence data storage means for storing correspondence between information elements in each of the plurality of information systems and data corresponding to the elements;
Using the stored contents of the element correspondence data storage means and the data of the element pairs with high structural consistency evaluated by the consistency evaluation means, data in the same field of different information sources, or logical operation of the data The information system association device according to attachment 3, further comprising data retrieval means for retrieving data corresponding to.

    (Additional remark 9) The said element pair detection means detects the element pair which adapts to this teacher data using the teacher data of the element pair designated from the outside among the elements which belong to a plurality of information systems The information system associating device according to appendix 1.

    (Supplementary note 10) The information system associating device according to supplementary note 1, wherein the information system is a category system as an information classification system, and the element is a category constituting the category system.

    (Supplementary note 11) The information system associating device according to supplementary note 10, wherein the data of the category is text data extracted from documents or text data as a meaningful character string.

(Supplementary note 12) The information system associating device according to supplementary note 10, wherein the data of the category is metadata including attribute data regarding an arbitrary classifiable object.
(Supplementary note 13) The information system associating device according to supplementary note 1, wherein the information system is a tag system corresponding to a structured document with a tag, and the element is a tag constituting the tag system.

(Additional remark 14) The said information system is a database table, The said element is a field of this database table, The information system matching apparatus of Additional remark 1 characterized by the above-mentioned.
(Supplementary Note 15) In an information system association method for examining matching between systems for a plurality of information systems,
Based on sample data corresponding to information element data belonging to multiple information systems, analyze the statistical characteristics of the data of individual information elements belonging to each information system,
Based on the analysis result, a common space for comparing a plurality of information systems is provided, and elements having similar statistical characteristics of the data of the elements among information elements belonging to different information systems on the common space Is detected as an element pair.

(Supplementary Note 16) In a program executed by a computer for examining matching between systems for a plurality of information systems,
A procedure for analyzing statistical characteristics of data of individual information elements belonging to each information system based on sample data corresponding to data of information elements belonging to a plurality of information systems;
Based on the analysis result, a common space for comparing a plurality of information systems is provided, and elements having similar statistical characteristics of the data of the elements among information elements belonging to different information systems on the common space For causing a computer to execute a procedure for detecting an element as an element pair.

(Supplementary Note 17) In a storage medium used by a computer for examining matching between systems for a plurality of information systems,
Analyzing statistical characteristics of data of individual information elements belonging to each information system based on sample data corresponding to data of information elements belonging to a plurality of information systems;
Based on the analysis result, a common space for comparing a plurality of information systems is provided, and elements having similar statistical characteristics of the data of the elements among information elements belonging to different information systems on the common space A computer-readable portable storage medium storing a program for causing a computer to execute the step of detecting an element pair.

  The present invention can be used in any industry that uses a large amount of data constituting a system.

It is a principle structure block diagram of the information system matching apparatus of this invention. It is a figure explaining the matching of the category in a heterogeneous classification system. It is a block diagram which shows the structure of the information system matching apparatus in a 1st Example. It is a whole flowchart of the category matching process in a 1st Example. It is explanatory drawing of the comparison of the category feature vector between systems. It is a detailed flowchart of the similar category pair detection process by a vector similarity. It is a figure explaining the similarity of a character string level. It is a figure which shows the structural example of a synonym synonym dictionary. It is explanatory drawing of the similarity determination method by a synonym synonym dictionary. It is a flowchart of a category name similarity determination process. It is a figure explaining the consistency of the hierarchical relationship in a heterogeneous classification system. It is a figure explaining the compatibility in the hierarchical relationship of two category pairs. It is a figure explaining the consistency of the neighborhood relationship in a heterogeneous classification system. It is a figure which shows the value of the proximity | contact relation matching degree corresponding to link distance. It is a detailed flowchart of an optimal category pair set detection process. It is a block diagram which shows the structure of the information system matching apparatus corresponding to a data search process. It is a figure which shows the data structure of a common category table. It is a figure which shows the data structure of a document-category index table. It is a figure which shows the data structure of a word-category index table. It is explanatory drawing of matching of the tag in a different tag system as a 2nd Example. It is explanatory drawing of the matching of the field in a different database as a 3rd Example. It is a figure explaining the loading to the computer of the program in this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Information system matching apparatus 2 Feature analysis means 3 Element pair detection means 4 Name similarity detection means 5 Consistency evaluation means 10 Control part 11 Information storage part according to category 12 Information hierarchy relation storage part 13 Category feature processing part 14 Category feature vector Storage unit 15 Category pair storage unit 16 Vector similarity processing unit 17 Category name similarity processing unit 18 Hierarchical relationship consistency processing unit 20 Common category table storage unit 21 Document-category index storage unit 22 Word-category index storage unit 23 Index creation Section 24 Search request processing section 25 Category level search section 26 Search result storage section 27 Search result display section 30 Central processing unit (CPU)
31 Read-only memory (ROM)
32 Random access memory (RAM)
33 Communication Interface 34 Storage Device 35 Input / Output Device 36 Reading Device 37 Bus 38 Program Provider 39 Network 40 Portable Storage Medium

Claims (10)

In an information system associating device for examining matching between systems for multiple information systems,
Feature analysis means for analyzing statistical characteristics of data of individual information elements belonging to each information system based on sample data corresponding to data of information elements belonging to a plurality of information systems;
Based on the analysis result, a common space for comparing a plurality of information systems is provided, and elements having similar statistical characteristics of the data of the elements among information elements belonging to different information systems on the common space An information system associating device, comprising: an element pair detecting means for detecting an element pair as an element pair. The information system associating device further includes name similarity detection means for detecting similarity of element names between information elements belonging to different information systems,
The element pair detection unit detects an element pair having a high integrated similarity by integrating the similarity of statistical characteristics of the data of the element and the similarity of the names. Information system association device.   In the information system associating device, the position in the information system of the element constituting the element pair detected by the element pair detecting means is mutually different from the position in the system of the element constituting the other element pair detected. 2. The information system associating device according to claim 1, further comprising consistency evaluating means for evaluating structural consistency indicating whether or not the information is consistent.   Among the plurality of information systems showing the directed graph relationship, the consistency evaluation means is a high order in the system of the elements constituting the detected element pair and the elements constituting the other detected element pair. 4. The information system associating device according to claim 3, wherein consistency of a hierarchical relation including a subordinate relation and / or a distance between elements is evaluated as the structural consistency.   The consistency evaluation means includes a distance between an element constituting the detected element pair and an element constituting the other detected element pair, among a plurality of information systems indicating undirected graph relationships. 4. The information system associating device according to claim 3, wherein consistency of neighboring relations is evaluated as the structural consistency. In the information system associating device,
Element correspondence data storage means for storing correspondence between information elements in each of the plurality of information systems and data corresponding to the elements;
Using the stored contents of the element correspondence data storage means and the data of the element pairs with high structural consistency evaluated by the consistency evaluation means, data in the same field of different information sources, or logical operation of the data 4. The information system associating device according to claim 3, further comprising data search means for searching for data corresponding to.   The element pair detection means detects an element pair that matches the teacher data by using teacher data of an element pair designated from the outside among elements belonging to a plurality of information systems. 1. The information system associating device according to 1. In the information system matching method for examining matching between systems for multiple information systems,
Based on sample data corresponding to information element data belonging to multiple information systems, analyze the statistical characteristics of the data of individual information elements belonging to each information system,
Based on the analysis result, a common space for comparing a plurality of information systems is provided, and elements having similar statistical characteristics of the data of the elements among information elements belonging to different information systems on the common space Is detected as an element pair. In a program executed by a computer that examines matching between systems for multiple information systems,
A procedure for analyzing statistical characteristics of data of individual information elements belonging to each information system based on sample data corresponding to data of information elements belonging to a plurality of information systems;
Based on the analysis result, a common space for comparing a plurality of information systems is provided, and elements having similar statistical characteristics of the data of the elements among information elements belonging to different information systems on the common space For causing a computer to execute a procedure for detecting an element as an element pair. In a storage medium used by a computer that examines matching between systems for multiple information systems,
Analyzing statistical characteristics of data of individual information elements belonging to each information system based on sample data corresponding to data of information elements belonging to a plurality of information systems;
Based on the analysis result, a common space for comparing a plurality of information systems is provided, and elements having similar statistical characteristics of the data of the elements among information elements belonging to different information systems on the common space A computer-readable portable storage medium storing a program for causing a computer to execute the step of detecting an element pair.

Images