JP2005157823A - Knowledge base system, inter-word meaning relation determination method in the same system and computer program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To uniquely specify an inter-concept path, and to reduce an operation load at the time of applying data showing meaning structures to an electronic document. <P>SOLUTION: This document classification processing part 2 operates the morphemic analysis of an electronic document fetched through a document input interface 11, and retrieves the carrying position of a concept dictionary for morphemes obtained by the morphemic analysis, and selects two morphemes from among the morphemes, and extracts the path of a conceptual relation line between the selected two morphemes, and extracts the proper conceptual relation line path according to weighting factors applied according to the strength of the inter-concept connection when a plurality of extracted paths are confirmed. Then, the conceptual relation data following the conceptual relation path are generated, and outputted through a document output interface 12 to the outside. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to a knowledge base system that captures an electronic document, classifies the electronic document based on the meaning and concept of the electronic document, and generates an output that matches the meaning and concept, and a method for determining a semantic relationship between words in the system , And its computer program.

  Many methods have been proposed for inferring the semantic content of electronic documents using concept dictionaries (also called ontology dictionaries) that describe the relationships between words in terms of trees and networks. Yes. For example, in order to perform flexible information extraction according to the field of the document, the purpose of the user, and the request for the document in various target fields, it uses the semantic relationship between the vocabularies in the ontology dictionary and does not appear in the document. Expand the word relationship sequence for the central word that briefly shows the user's interest while supplementing the relationship, and attribute the individual word relationship sequence based on the partial word relationship and how the entire word is used. And a method of converting the relationship into attribute values has been proposed (see Patent Document 1).

  In addition, in order to efficiently provide various services to users, when providing content in response to inquiries in natural language from users, the ontology dictionary is used to conceptualize knowledge about the content and There has also been proposed a method of providing content to a user by providing interactive support for an inquiry (see Patent Document 2).

Furthermore, in order to obtain a vocabulary acquisition method that can acquire parts of speech and vocabulary concepts even for words that are not registered in the dictionary, by referring to the basic word dictionary and the term dictionary, the morphemes of the input text Analyze, and then refer to the term pattern dictionary to match the words that appear in the morphological analysis result with the words that are listed in the term pattern dictionary. A method of extracting a term even when a word appearing in the result of morphological analysis is a subordinate concept of a word concept described in a term pattern dictionary has been proposed (see Patent Document 3). ). Here, “morpheme” refers to a minimum character string that has no meaning if it is further reduced, and “morpheme analysis” refers to breaking a sentence into morphemes.
Japanese Unexamined Patent Publication No. 2000-207407 (paragraphs 0007 to 0011, FIG. 1) JP 2002-197107 A (paragraphs 0008 and 0009, FIG. 1) JP 2001-67356 A (paragraphs 0007 to 0019, FIG. 1)

  Many of the above-mentioned patent documents use a tree-structured concept dictionary describing the relationship between upper and lower concepts (so-called has-a relationship, is-a relationship) between nouns in order to infer the semantic content of a document. ing. For example, as shown in FIG. 7 which shows the conceptual relationship in the conventional tree-structured concept dictionary, the path (thick line in the figure) connecting the two concepts A and B is uniquely determined without using a special logical method. be able to.

  However, the conceptual system is not a simple tree structure, but generally has a network structure (network) as shown in FIG. That is, in addition to the upper and lower concept connections, there are conceptual connections such as equivalence, contradiction, object relationship, and causal relationship, and these have a correspondence relationship different from the directionality and inclusion relationship such as upper and lower. That is, in FIG. 8, concept C has two superordinate concepts of concept E and concept F, concept D has two superordinate concepts of concept F and concept G, and concept C and concept D are superordinate and subordinate. It has a conceptual relationship such as equivalence.

  The difference between the tree structure and the network structure is whether the superordinate concept is uniquely determined or a plurality of concepts can exist. In a concept dictionary of network structure, there may be multiple superordinate concepts for one concept or other conceptual relationships, so there are multiple paths of concept relationship lines connecting two concepts. Also occurs. For example, in the example shown in FIG. 9, there are paths of concept relationship lines indicated by solid lines and alternate long and short dash lines between Concept H and Concept D, and other paths can be taken. Here, a connection between concepts including upper and lower concepts is called a concept relationship, and a line indicating the connection is called a concept relationship line. In this case, the path between the two concepts cannot be uniquely specified by any of the above patent documents.

  On the other hand, as one method for expressing the semantic content of an electronic document, there is a description by metadata that is information for explaining the content of information (content) such as text, image, and sound. Conventionally, it has been left to human naval tactics to give metadata to electronic documents. Therefore, a great burden has been imposed on workers (document registrants), and improvement by mechanization has been desired.

  The present invention has been made in view of the above-mentioned circumstances, and uses a network-structured concept dictionary that covers conceptual relationships in linguistics, and automatically discriminates the semantic structure of an electronic document in a form that conforms to a natural language. This makes it possible to uniquely identify the path between concepts, reduce the work load when assigning data indicating a semantic structure to an electronic document, and cause the electronic document to be buried due to omissions in search, etc. It is an object of the present invention to provide a knowledge base system, a method for determining semantic relations between words in the system, and a computer program thereof that can be used effectively without any problems. Here, the “semantic structure” refers to a structure showing a semantic connection between morphemes obtained by semantic analysis (described later).

  In order to solve the above-described problems, the present invention provides a knowledge base system including an arithmetic unit that takes in an electronic document, classifies it based on the meaning and concept of the electronic document, and generates an output that matches the meaning and concept When a plurality of relationships are confirmed by referring to a concept dictionary for a conceptual relationship between morphemes obtained by performing a morphological analysis of the electronic document and a document input interface unit that captures the electronic document and the electronic document. A document classification processing unit constructed in the arithmetic unit for generating data representing the conceptual relationship between the morphemes according to a weighting factor given according to the strength of the conceptual relationship, and data representing the conceptual relationship between the morphemes And a document output interface unit for outputting.

  In the present invention, the document classification processing unit includes a morpheme analysis unit that decomposes the electronic document into morphemes by morpheme analysis, a morpheme placement position analysis unit that searches for a placement position of the concept dictionary for the morpheme, Select any two of the morphemes, extract the path of the conceptual relationship line between the two morphemes, and when multiple extracted paths are confirmed, depending on the strength of the connection between the concepts A conceptual relationship analysis unit that extracts a path of a concept relationship line according to a given weighting factor and a concept relationship data generation unit that generates data of a concept relationship according to the path of the concept relationship line are provided.

  Further, in the present invention, the concept relationship analysis unit calculates a reciprocal of the weighting factor to obtain a distance of the concept relationship line, calculates a sum of distances along a path connecting the two concepts of interest, The path having the minimum sum is extracted as the path of the concept relation line.

  Further, in the present invention, a syntax analysis unit that analyzes a grammatical structural relationship of a sentence to select a combination of morphemes from which the concept relationship is to be extracted, and supplies the combination to the concept relationship analysis unit It is further provided with the feature.

  In order to solve the above-described problems, the present invention captures an electronic document, classifies it based on the meaning and concept of the electronic document, and generates an output that matches the meaning and concept, meaning between words A relationship determination method, comprising: performing a morpheme analysis of the electronic document; searching a posting position of the concept dictionary for a morpheme obtained by the morpheme analysis; and selecting any two of the morphemes Then, the path of the concept relationship line between the two morphemes is extracted, and when a plurality of the extracted paths are confirmed, the concept relationship line is determined according to the weighting coefficient assigned according to the strength of the connection between the concepts. The method includes a step of extracting a path and a step of generating data on a concept relationship according to the path of the concept relationship line.

  In order to solve the above problems, the present invention is a computer used in a knowledge base system that takes in an electronic document, classifies it based on the meaning and concept of the electronic document, and generates a search result that matches the meaning and concept. A program for performing a morphological analysis of the electronic document, a process for searching for a posting position of the concept dictionary for a morpheme obtained by the morpheme analysis, and selecting any two of the morphemes, The path of the concept relation line between the two morphemes is extracted, and when a plurality of the extracted paths are confirmed, the path of the concept relation line is determined according to the weighting coefficient given according to the strength of the connection between the concepts. And causing the computer to execute a process of extracting and a process of generating data on a conceptual relationship according to a path of the conceptual relationship line and storing the data in the storage device That.

  According to the present invention, when the document classification processing unit performs morphological analysis of an electronic document and a plurality of relationships are confirmed by referring to the concept dictionary regarding the conceptual relationship between morphemes obtained by the morpheme analysis, By generating data that expresses the conceptual relationship between morphemes according to the weighting factor assigned according to the strength of the relationship, the semantic structure of the electronic document can be automatically identified in a more natural language form. This makes it possible to uniquely identify the path between concepts, reduce the work load when assigning data indicating the semantic structure to the electronic document, and bury the electronic document due to omission of search, etc. It is possible to provide a knowledge base system that can be used effectively without any problems.

  Further, according to the present invention, when the concept relationship analysis unit constituting the document classification processing unit extracts the path of the concept relationship line between the two selected morphemes, and when a plurality of the extracted paths are confirmed The path of the concept relationship line is extracted according to the weighting factor assigned based on the co-occurrence relationship, for example, according to the strength of the connection between the concepts, and the concept relationship data generation unit extracts the path of the concept relationship line By generating the conceptual relationship data according to the above, it is possible to reduce the work load when the data indicating the semantic structure is given to the electronic document. Here, “co-occurrence” means that a certain word and another word are included in the same document, and “co-occurrence relationship” means a certain word determined by the frequency of co-occurrence described above. The degree of perspective with other words.

  Further, according to the present invention, the concept relation analysis unit obtains the distance by calculating the reciprocal of the weighting factor, and further calculates the sum of the distances along the path connecting the two target concepts. By extracting the path with the smallest value, it can be extracted as the path of the conceptual relationship line, and this allows the semantic structure of the electronic document to be automatically identified in a form that complies with natural language. Data indicating a semantic structure can be easily given to a document.

  In addition, according to the present invention, it is possible to select only morphemes having a dependency relationship by selecting a combination of morphemes for which a conceptual relationship is to be extracted by performing a syntax analysis. It is possible to reduce the processing load on the computer for finding the conceptual relationship.

  FIG. 1 is a block diagram showing an embodiment of the knowledge base system of the present invention. The knowledge base system of the present invention performs morphological analysis of an electronic document, and when a plurality of relationships are confirmed with reference to a concept dictionary regarding the concept relationship between morphemes obtained by the morpheme analysis, the strength of the relationship between concepts A document classification processing unit 2 constructed on an arithmetic unit that generates data representing a conceptual relationship between morphemes according to a weighting factor assigned according to the document, and a document input interface unit 11 for capturing an electronic document, or a concept between morphemes It is connected to an external system via a document output interface unit 12 that outputs data representing the relationship.

  The document classification processing unit 2 further includes a morpheme analysis unit 21, a morpheme placement position analysis unit 22, a syntax analysis unit 23, a concept relationship analysis unit 24, a concept relationship data generation unit 25, a morpheme dictionary 26, a concept dictionary 27, and concept relationship data. It is composed of a storage DB (Data Base) 28.

  The morpheme analysis unit 21 has a function of referring to the morpheme dictionary 26 for an electronic document captured via the document input interface unit 11, decomposing the electronic document into morphemes, and supplying the morpheme placement position analysis unit 22.

  Further, the morpheme placement position analysis unit 22 searches for the placement position of the concept dictionary 27 for the morpheme supplied from the morpheme analysis unit 21 and supplies it to the syntax analysis unit 23, and the syntax analysis unit 23 is omitted. In some cases, it has a function of supplying to the conceptual relationship analysis unit 24.

  On the other hand, the syntax analysis unit 23 has a function of performing a well-known syntax analysis to select a combination of morphemes from which a concept relationship is to be extracted, and supplying the combination to the concept relationship analysis unit 24. It is not prepared and may be omitted.

  The conceptual relationship analysis unit 24 selects any two of the morphemes analyzed by the morpheme analysis unit 21, extracts the path of the conceptual relationship line between the two selected morphemes, and the extracted path is When a plurality of the relations are confirmed, the path of the concept relation line is extracted according to the weighting coefficient assigned according to the strength of the connection between the concepts and supplied to the concept relation data generation unit 25. The conceptual relationship analysis unit 24 also calculates the reciprocal of the weighting factor to obtain the distance of the conceptual relationship line, calculates the sum of the distances along the path connecting the two target concepts, and the path that minimizes the sum It also has a function to extract as a path of conceptual relationship lines.

  The concept relationship data generation unit 25 has a function of generating concept relationship data according to the path of the concept relationship line and storing it in the concept relationship data storage DB 28.

  FIGS. 2 and 3 are flowcharts cited for explaining the operation of the knowledge base system in the present invention, and also show the processing procedure of the computer program in the present invention. FIG. 4 to FIG. 6 are operation concept diagrams quoted for the purpose of helping understanding of the operation of the knowledge base system according to the present invention. Examples of syntax analysis (FIG. 4) and connections between concepts in the concept dictionary 27 are shown. FIG. 5 shows an image (FIG. 5) and an example (FIG. 6) of path selection as a result of conceptual relationship analysis. The operation of the knowledge base system of the present invention shown in FIG. 1 will be described in detail below with reference to FIGS.

  First, as a precondition, the concept dictionary 27 used in the present invention holds a conceptual relationship in natural linguistics, and the conceptual relationship includes a weighting factor according to the strength of the relationship, for example, language It is assumed that weighting is based on academic “co-occurrence relationships”.

  First, an electronic document to be given semantic data is fetched from another system via the document input interface unit 11 (S201). Then, the morpheme analysis unit 21 refers to the morpheme dictionary 26 and decomposes the electronic document into morpheme units (minimum units of word structure), performs well-known morpheme analysis to clarify the properties of each morpheme, It supplies to the position analysis part 22 (S202). It is assumed that the electronic document is composed of one or a plurality of sentences.

  The morpheme placement position analysis unit 22 selects one of the sentences and searches for the placement position of the morpheme contained in the sentence on the concept dictionary 27 (S203).

  At this time, the syntax analysis unit 23 may select only morphemes having a dependency relationship. As is well known, syntactic analysis is to determine whether a morphologically analyzed sentence is a correct sentence using grammar, and when it is a correct sentence, a network structure is obtained as a result of the syntactic analysis. Semantic analysis is performed to select the correct path from these. Here, “semantic analysis” refers to extracting a semantic / conceptual form from a syntactically analyzed sentence and analyzing a conceptual structure of the sentence.

  FIG. 4 shows an example in which syntactic analysis is performed to select combinations of morphemes from which concept relationships are to be extracted. In other words, if the sentence to be parsed is “The Council General Assembly was held in Shimbashi”, the combination of morphemes that require semantic analysis is “Council”, “General Assembly”, and “General Assembly”. "Open", "Shinbashi" and "Open", for example, there is no syntactic connection for the combination of "Conference" and "Shimbashi", "General Assembly" and "Shinbashi", Even if analysis is not performed, the overall analysis result is not affected. Therefore, by performing syntax analysis here, it is possible to select combinations of morphemes to be semantically analyzed, and at this time, it is possible to reduce processing for finding a conceptual relationship between morphemes.

  Next, the conceptual relationship analysis unit 24 selects any two of the morphemes analyzed by the morpheme analysis unit 21 (S204), and from the posted positions on the concept dictionary 27 for the two selected morphemes, The path of the conceptual relationship line between the morphemes is extracted (S205).

  FIG. 5 shows a connection image between concepts in the concept dictionary 27. Here, the terms (higher and lower ranks, principals, and objects) shown accompanying the concept relationship line indicate the type of concept relationship, and the numerical value indicates the weighting coefficient of the concept relationship. As described above, the connection between concepts in the concept dictionary 27, that is, the concept relationship line, is given a weighting factor corresponding to the strength of the relationship. Specifically, there is weighting based on the co-occurrence relationship. A co-occurrence relationship means a relationship between words that appear simultaneously in one sentence. The co-occurrence frequency may be the weight of the concept relationship line. Moreover, you may use the weight attached | subjected by linguistic conceptual relations other than co-occurrence relations, for example, the deep case relation which is the semantic relation which the verb in a sentence has with a noun concept.

  By the way, it is determined whether or not there are a plurality of extracted paths (S206). If there are a plurality of paths (Yes in S206), the conceptual relationship analysis unit 24 calculates the “distance” of each path according to the following procedure. Then, the smallest path is selected.

  Here, the reciprocal of the weight coefficient is defined as the “distance” of the concept relation line, the sum of the distances is taken along the path connecting the two target concepts, and the one that minimizes the sum is selected (S207). If there are a plurality of minimum items, all of them may be selected, or any one may be selected.

  FIG. 6 shows an example of path selection. Here, between the concept A-concept C, the concept C-concept D, and the concept D-concept B, the co-occurrence relationship weight coefficient is "2", and between the concept A-concept E, concept E-concept B is co-occurrence. The relationship weighting factor is “1”. Therefore, the sum of the weighting factors of the path A-C-D-B is 1/2 + 1/2 + 1/2 = 3/2, and the sum of the weighting factors of the path A-E-B is 1/1 + 1/1 = 2. Therefore, since the former is less than the latter, the path A-C-D-B is selected.

  When there is a single path (No in S206), the process of S207 described above is skipped, and the process of S208 described later is executed. In S208, the concept relation data generation unit 25 generates data representing the concept relation for the morpheme and stores it in the concept relation data storage DB. Here, as a specific example of the data format, there is RDF (Resource Description Framework) format data which is a description method standardized by W3C (World Wide Web Consortium). Of course, it is not limited to the RDF format.

  The series of processing (S205 to S208) by the conceptual relationship analysis unit 24 described above is repeated for all morpheme sets (or morpheme sets in a dependency relationship) included in the sentence (S209). Then, all the conceptual relationship data extracted for the sentence is collected together (S210), and the above-described series of processing (S203 to S210) is repeated for all sentences included in the electronic document (S211).

  Then, after completing all the analysis processes, the conceptual relationship analysis unit 24 activates the conceptual relationship data generation unit 25, and the conceptual relationship data generation unit 25 generates data representing the conceptual relationship with respect to the morpheme. Then, the data is output to an external system (not shown) via the document output interface 12 (S212).

  As described above, the present invention makes it possible to construct a knowledge base system that can share, distribute, or search information including not only superficial character information of an electronic document but also its semantic content. It is. Further, since it is possible to automatically determine the semantic content from the electronic document and assign the result as semantic data, the burden on the operator who performs document registration is greatly reduced.

  Furthermore, by using a network-structured concept dictionary that covers conceptual relationships in linguistics, it is possible to absorb fluctuations of words due to differences in document creators and to provide semantic data with semantic uniformity. Become. This uniformity of meaning makes it possible to improve search performance and automatically classify documents.

  The document classification processing unit 2 shown in FIG. 1 and the morpheme analysis unit 21, the morpheme placement position analysis unit 22, the syntax analysis unit 23, the concept relationship analysis unit 24, and the concept relationship data generation unit that constitute the document classification processing unit 2 The knowledge base system of the present invention is also realized by recording the procedure executed in each of the programs 25 on a computer-readable recording medium, causing the computer system to read and execute the program recorded on the recording medium. Can do. The computer system here includes hardware such as an OS (Operating System) and peripheral devices.

It is a block diagram which shows one Embodiment of the knowledge base system of this invention. It is the flowchart quoted in order to demonstrate operation | movement of the knowledge base system in this invention. It is the flowchart quoted in order to demonstrate operation | movement of the knowledge base system in this invention. It is the operation | movement conceptual diagram quoted in order to assist the understanding of operation | movement of the knowledge base system in this invention. It is the operation | movement conceptual diagram quoted in order to assist the understanding of operation | movement of the knowledge base system in this invention. It is the operation | movement conceptual diagram quoted in order to assist the understanding of operation | movement of the knowledge base system in this invention. It is the figure quoted in order to demonstrate the conceptual relationship in the conventional tree-structured concept dictionary. It is the figure quoted in order to demonstrate the conceptual relationship in the conventional network structure concept dictionary. It is the figure quoted in order to demonstrate the path | route of the concept relation line in the conventional network structure concept dictionary.

Explanation of symbols

DESCRIPTION OF SYMBOLS 11 Document input interface part 12 Document output interface part 2 Document classification | category process part 21 Morphological analysis part 22 Morphological placement position analysis part 23 Syntax analysis part 24 Concept relation analysis part 25 Concept relation data generation part 26 Morphological dictionary 27 Concept dictionary 28 Concept relation data Save DB

Claims (6)

A knowledge base system including an arithmetic device that takes in an electronic document, classifies it based on the meaning and concept of the electronic document, and generates an output that matches the meaning and concept,
A document input interface unit for capturing the electronic document;
When the morpheme analysis of the electronic document is performed and a plurality of relationships are confirmed with reference to the concept dictionary stored in the storage device, the strength of the concept relationship is determined. A document classification processing unit constructed in the arithmetic device for generating data representing a conceptual relationship between the morphemes according to a weighting factor assigned accordingly,
A knowledge base system comprising: a document output interface unit that outputs data representing a conceptual relationship between the morphemes. The document classification processing unit
A morpheme analyzer that decomposes the electronic document into morphemes by morpheme analysis;
A morpheme placement position analysis unit for searching for a placement position of the concept dictionary for the morpheme;
Select any two of the morphemes, extract the path of the conceptual relationship line between the two morphemes, and depending on the strength of the connection between the concepts when a plurality of the extracted paths are confirmed A concept relationship analysis unit that extracts a path of a concept relationship line according to the weighting factor assigned in
The knowledge base system according to claim 1, further comprising a concept relation data generation unit that generates data of a concept relation according to a path of the concept relation line. The conceptual relationship analysis unit
The reciprocal of the weight coefficient is calculated as the distance of the conceptual relationship line, the sum of distances is calculated along the path connecting the two target concepts, and the path having the minimum sum is defined as the path of the conceptual relationship line. The knowledge base system according to claim 2, wherein the knowledge base system is extracted.   A syntactic analysis unit that selects a combination of morphemes for which the conceptual relationship is to be extracted by performing a syntactic analysis and supplies the combination to the conceptual relationship analysis unit or the morpheme placement position analysis unit; The knowledge base system according to claim 2 or 3. The method of determining a semantic relationship between words in a knowledge base system according to claim 1, wherein an electronic document is taken in, classified based on the meaning and concept of the electronic document, and an output that matches the meaning and concept is generated. ,
Performing a morphological analysis of the electronic document;
Searching the posting position of the concept dictionary for morphemes obtained by the morpheme analysis;
Select any two of the morphemes, extract the path of the conceptual relationship line between the two morphemes, and depending on the strength of the connection between the concepts when a plurality of the extracted paths are confirmed Extracting a path of the concept relation line according to the weighting factor given in
Generating semantic relationship data according to the path of the conceptual relationship line, and determining a semantic relationship between words in a knowledge base system. A computer program used in the knowledge base system according to claim 1, which takes in an electronic document, classifies it based on the meaning and concept of the electronic document, and generates an output that matches the meaning and concept,
Processing to perform morphological analysis of the electronic document;
A process of searching for a posting position of the concept dictionary for a morpheme obtained by the morpheme analysis;
Select any two of the morphemes, extract the path of the conceptual relationship line between the two morphemes, and depending on the strength of the connection between the concepts when a plurality of the extracted paths are confirmed A process of extracting the path of the concept relation line according to the weighting factor given in
The computer program which makes a computer perform the process which produces | generates the data of the conceptual relationship according to the path | route of the said conceptual relationship line.

Images