JP2010224833A - Ontology generation device and method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an ontology generation device and method, which allows a user to easily determine inconsistency between concepts constituting a generated ontology. <P>SOLUTION: An acquisition unit 45 obtains document data. A first extraction unit 50 extracts a pattern showing a dependency between a first and second character strings in which words showing a concept pair that is a set of concepts stored in a concept information storage part 31 are replaced by variables, of character strings of a sentence in which the concept pair is co-occurring, with other character strings from the document data and stores it in a pattern information storage part 36. A second extraction unit 55 extracts a new concept pair from the document data using the pattern stored in the pattern information storage part 36, and stores it in the concept information storage part 31. A generation unit 60 generates an ontology using a plurality of concept pairs stored in the concept information storage part 31, and a determination unit 65 determines presence/absence of inconsistency between concepts constituting the ontology, and an output unit 20 outputs the determination result together with the ontology. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates to an ontology generation apparatus and method.

  Conventionally, ontology techniques used for information processing apparatuses and the like are known. “Ontology” is a type of dictionary that systematically organizes the concepts of words. In ontology, the positional relationship of concepts expresses the relative meaning between concepts, and it has become possible to perform advanced knowledge processing by using this semantic information.

  In recent years, efforts have been made to develop an ontology by extracting concepts and relationships between concepts from a large amount of document data.

  As such a technique, natural language processing technology is applied to document data, and the concept and relationship are extracted using part-of-speech information, syntax information, or semantic information (for example, a character string pattern or a syntax tree pattern), There are techniques to build ontology. In addition, feature language and sentence linguistic information (part of speech information, syntax information, semantic information, etc.) obtained by applying natural language processing technology to document data are used as features to extract a set of concepts that have the correct relationship. There are techniques for constructing an ontology (see Non-Patent Documents 1 and 2).

Patrick Pantel, Marco Pennacchioti, “Espresso: Leveraging Generic Patters for Automatic Coding and Costensive Relations”. 113-120, 2006 Fabian M.M. Suchanek, Georgiana Ifrim, Gerhard Weikum, "LEILA: Learning to Extract Information by Linguistic Analysis and Wrong and Plow. 18-25, 2006

  However, in the conventional technology as described above, there is a possibility that the extracted concept set includes a set of concepts that have contradictions such as mutual reference to each other although the relationship between the concepts is correct. There is. The user may not be able to determine that such a contradiction has occurred.

  The present invention has been made in view of the above circumstances, and an object thereof is to provide an ontology generation apparatus and method that allow a user to easily discriminate between contradictions between the concepts constituting the generated ontology. .

  In order to solve the above-described problems and achieve the object, an ontology generation apparatus according to one aspect of the present invention includes a acquisition unit that acquires document data and a concept information storage that stores a concept pair that is a set of vocabulary concepts. And the first and second character strings obtained by replacing each vocabulary representing the concept of the concept pair with a variable among the character strings of the sentence in which the concept pair co-occurs from the document data and other character strings. A first extraction unit that extracts a pattern indicating a dependence relationship, a pattern information storage unit that stores the extracted pattern, and a new information from the document data using the pattern stored in the pattern information storage unit. An ontology in which relationships between concepts are organized using a second extraction unit that extracts concept pairs and stores them in the concept information storage unit, and a plurality of concept pairs stored in the concept information storage unit A generating unit that generates a determination unit that determines presence or absence of conflict between concepts that constitute the ontology, characterized in that it comprises an output unit for outputting a determination result together with the ontology.

  An ontology generation method according to another aspect of the present invention includes an acquisition step in which an acquisition unit acquires document data, and a first extraction unit that stores a concept pair that is a set of vocabulary concepts from the document data. First and second character strings and other character strings obtained by replacing each vocabulary representing the concept of the concept pair with a variable among character strings of sentences co-occurring with the concept pair stored in the concept information storage unit A first extraction step of extracting a pattern indicating a dependency relationship with the pattern information storage unit and storing the pattern in the pattern information storage unit, and the second extraction unit using the pattern stored in the pattern information storage unit, the document data A second extraction step of extracting a new concept pair from the concept information storage unit and storing it in the concept information storage unit, and a generation unit using the plurality of concept pairs stored in the concept information storage unit, The A generation step for generating the processed ontology, a determination step for determining whether there is a contradiction between the concepts constituting the ontology, and an output control step for causing the output unit to output a determination result together with the ontology. It is characterized by including these.

  According to the present invention, there is an effect that a user can easily discriminate between contradictions between the concepts constituting the generated ontology.

FIG. 1 is a diagram illustrating an example of an ontology. FIG. 2 is a block diagram illustrating an example of the configuration of the ontology generation apparatus according to the present embodiment. FIG. 3 is a diagram illustrating an example of concept information stored in the extracted concept information storage unit of the present embodiment. FIG. 4 is a diagram illustrating an example of concept information stored in the user concept information storage unit of the present embodiment. FIG. 5 is a diagram illustrating an example of pattern information stored in the extracted pattern information storage unit of the present embodiment. FIG. 6 is a diagram illustrating an example of pattern information stored in the user pattern information storage unit of the present embodiment. FIG. 7 is a diagram illustrating an example of syntax tree information according to the present embodiment. FIG. 8 is a diagram illustrating an example of a syntax tree pattern according to the present embodiment. FIG. 9 is a diagram illustrating an example of a state in which new pattern information is stored in the extracted pattern information storage unit of the present embodiment. FIG. 10 is a diagram illustrating an example of a syntax tree pattern according to the present embodiment. FIG. 11 is a diagram illustrating an example of syntax tree information of a sentence having a syntax tree pattern according to the present embodiment. FIG. 12 is a diagram illustrating an example of a state in which new concept information is stored in the extracted concept information storage unit of the present embodiment. FIG. 13 is a diagram illustrating an example of the ontology of the present embodiment. FIG. 14 is a diagram illustrating an example of an ontology output mode according to the present embodiment. FIG. 15 is a diagram illustrating an example of a state in which the contradiction determination result is stored in the extracted concept information storage unit of the present embodiment. FIG. 16 is a diagram illustrating an example of a state in which the contradiction determination result is stored in the user concept information storage unit of the present embodiment. FIG. 17 is a flowchart illustrating an example of a procedure flow of processing performed by the ontology generation device according to the present embodiment. FIG. 18 is a flowchart illustrating an example of a flow of a procedure of pattern information extraction processing according to the present embodiment. FIG. 19 is a diagram illustrating an example of a vector used for machine learning of a syntax tree pattern. FIG. 20 is a diagram illustrating an example of a concept pair and a sentence in which the concept pair co-occurs. FIG. 21 is a diagram illustrating an example of a vector used for machine learning of a syntax tree pattern. FIG. 22 is a diagram illustrating an example of the concept of the classifier according to the present embodiment. FIG. 23 is a diagram illustrating an example of a classifier according to the present embodiment. FIG. 24 is a flowchart illustrating an example of a flow of a concept information extraction process according to the present embodiment. FIG. 25 is a diagram illustrating an example of testing the correlation between concept pairs. FIG. 26 is a flowchart illustrating an example of the procedure of ontology generation processing according to the present embodiment. FIG. 27 is a flowchart illustrating an example of the flow of the contradiction determination process according to the present embodiment. FIG. 28 is a flowchart illustrating an example of a flow of a registration process according to the present embodiment. FIG. 29 is a flowchart illustrating an example of a flow of comparison processing according to the present embodiment.

  DESCRIPTION OF EXEMPLARY EMBODIMENTS Hereinafter, embodiments of an ontology generation device and method according to the invention will be described in detail with reference to the accompanying drawings.

  First, an overview of the ontology will be described.

  "Ontology" is a systematic arrangement of the concept of words. As the “concept”, two types of classes and instances are mainly used, but the present invention is not limited to this. “Class” indicates the classification name of the concept, and “Instance” indicates an example of the concept.

  In Ontology, the positional relationship of concepts placed on the ontology expresses the relative meaning between the concepts. In general, the relationship between concepts is an upper-lower relationship (is-a relationship), the whole Any of a partial relationship (part-of relationship) and an instance relationship (instance-of relationship) is included.

  FIG. 1 is a diagram illustrating an example of an ontology. In the example shown in FIG. 1, the ontology is composed of seven classes and three instances. The class indicates the vehicle type classification of the automobile, and the instance indicates the year of the automobile. In the example shown in FIG. 1, the relationship between classes is an upper-lower relationship, and the relationship between classes and instances is an instance relationship.

  For example, the ontology description language OWL, which is an ontology description language, can be used, but is not limited thereto.

  Next, the configuration of the ontology generation device according to the present embodiment will be described.

  FIG. 2 is a block diagram illustrating an example of the configuration of the ontology generation device 1 according to the present embodiment. As shown in FIG. 2, the ontology generation device 1 includes an input unit 10, an output unit 20, a storage unit 30, a reception unit 40, an acquisition unit 45, a pattern extraction unit 50, a concept extraction unit 55, A generation unit 60, a determination unit 65, an output control unit 70, a registration unit 75, and a comparison unit 80 are provided.

  The input unit 10 performs input of various operations such as an acquisition operation instructing acquisition of a document, and can be realized by an existing input device such as a keyboard, a mouse, or a touch panel.

  The output unit 20 outputs an ontology generated by a generation unit 60 described later, a determination result of a determination unit 65 described later, and the like according to an instruction from an output control unit 70 described later. For example, a CRT display or a liquid crystal display It can be realized by an existing display device such as a plasma display, an organic EL display, or a touch panel display. The output unit 20 may be realized by an existing printing apparatus such as a printer, or may be realized by using these together.

  The storage unit 30 stores information used for various processes performed by the ontology generation device 1. For example, an HDD (Hard Disk Drive), an SSD (Solid State Drive), a memory card, an optical disk, or a RAM ( It can be realized by an existing storage medium that can store magnetically, electrically, or optically such as Random Access Memory. The storage unit 30 includes a concept information storage unit 31 and a pattern information storage unit 36.

  The concept information storage unit 31 stores concept information including a concept pair that is a set of concepts, and an extracted concept information storage that stores concept information including a concept pair extracted by a concept extraction unit 55 described later. Unit 32 and a user concept information storage unit 33 for storing concept information registered by the user.

  FIG. 3 is a diagram illustrating an example of concept information stored in the extracted concept information storage unit 32, and FIG. 4 is a diagram illustrating an example of concept information stored in the user concept information storage unit 33. In the example illustrated in FIGS. 3 and 4, the concept information is information including a concept pair (concept 1 and concept 2), a relationship, a label (an example of a first label), and a contradiction. The “concept pair” is information indicating a vocabulary pair representing a concept that is a component of the ontology. “Relationship” is information indicating the relationship between concept pairs. “Label” is information indicating whether or not the relationship between concept pairs is correct, and is “positive example” if the relationship is correct and “negative example” if the relationship is not correct. “Contradiction” is information indicating whether or not there is a contradiction such as a cross-reference or a cyclic reference between concept pairs. If there is a contradiction, “Yes” is indicated, and if there is no contradiction, “No” is indicated.

  The pattern information storage unit 36 stores pattern information including a syntax tree pattern indicating the dependency relationship between character strings, and stores pattern information including a syntax tree pattern extracted by the pattern extraction unit 50 described later. An extraction pattern information storage unit 37 and a user pattern information storage unit 38 that stores pattern information registered by the user are included.

  FIG. 5 is a diagram illustrating an example of pattern information stored in the extracted pattern information storage unit 37, and FIG. 6 is a diagram illustrating an example of pattern information stored in the user pattern information storage unit 38. In the example shown in FIGS. 5 and 6, the pattern information is information including a syntax tree pattern, a relationship, and a label (an example of a second label). The “syntax tree pattern” is information indicating the dependency relationship of character strings. “Relationship” is information indicating the relationship between concept pairs extracted by the syntax tree pattern. “Label” is information indicating whether or not the concept pair is valid for extraction, and is “positive example” if valid and “negative example” if not valid.

  The accepting unit 40 accepts input of various operations input by the input unit 10. Specifically, the receiving unit 40 performs an acquisition operation, a concept information registration operation with respect to the user concept information storage unit 33, a pattern information registration operation with respect to the user pattern information storage unit 38, an extraction concept information storage unit 32, and a user concept information storage. A conceptual information comparison operation between the units 33 or a pattern information comparison operation between the extracted pattern information storage unit 37 and the user pattern information storage unit 38 is accepted.

  The acquisition unit 45 acquires document data. Specifically, the acquisition unit 45 acquires document data when an acquisition operation is received by the reception unit 40. The document data may be stored in the storage unit 30, and the acquisition unit 45 may acquire the document data from the storage unit 30. For example, the document data stored in an external device (not shown) such as a server may be stored. You may make it acquire.

  The document data acquired by the acquisition unit 45 is a collection of document data written for a specific domain, and examples thereof include patent documents, technical documents, sales documents, and business documents. The format of the document data acquired by the acquisition unit 45 is not limited, and may be any of a structured document, a semi-structured document, and an unstructured document.

  The pattern extraction unit 50 (an example of a first extraction unit) uses the concept from the document data acquired by the acquisition unit 45 among the character strings of the sentences in which the concept pairs stored in the concept information storage unit 31 co-occur. A pattern indicating a dependency relationship between the first and second character strings in which each vocabulary representing the concept of the pair is replaced with a variable and another character string is extracted.

  Specifically, the pattern extraction unit 50 reads the concept information from the extracted concept information storage unit 32 or the user concept information storage unit 33, and searches for document data in which the concept pairs of the read concept information co-occur. Then, the pattern extraction unit 50 parses the retrieved document data, and extracts sentences in which concept pairs co-occur as syntax tree information.

  FIG. 7 is a diagram illustrating an example of syntax tree information extracted by the pattern extraction unit 50 using the conceptual information 113 illustrated in FIG. 4. The syntax tree information shown in FIG. 7 is syntax tree information generated by parsing a sentence composed of character strings 121 to 125, and the character strings 121 to 125 are syntax trees. And the syntax tree information shown in FIG. 7 has shown the dependency relationship of each of the character strings 121-125 which are syntax trees.

  Then, the pattern extraction unit 50 deletes an unnecessary character string from the extracted syntax tree information and replaces the character string corresponding to the concept information with a variable in the extracted concept information storage unit 32 together with the relationship and the label. Remember.

  In addition, as deletion of an unnecessary character string, the pattern extraction part 50 deletes character strings other than the shortest path | pass where the character string corresponding to conceptual information appears, and the said character string are connected directly. The relationship stored in the extracted concept information storage unit 32 is information indicating the relationship between concept pairs used for extracting the syntax tree pattern, and the labels stored in the extracted concept information storage unit 32 will be described later.

  FIG. 8 is a diagram illustrating an example of the syntax tree pattern of the syntax tree information illustrated in FIG. In the example shown in the figure, the character string 121 is deleted from the syntax tree information shown in FIG. 7, and the character strings 122 and 124 corresponding to the concept pairs in the concept information 113 are replaced with character strings 132 and 134 including variables, respectively. Yes.

  FIG. 9 is a diagram illustrating an example of a state in which pattern information including the syntax tree pattern illustrated in FIG. 8 is newly stored in the extracted pattern information storage unit 37. In the example shown in FIG. 9, a syntax tree pattern composed of character strings 132, 123, 134, and 125 is stored in the extracted pattern information storage unit 37 as pattern information 141 together with relationships and labels. Note that the conceptual information used for extracting the syntactic tree pattern shown in FIG. 8 is the conceptual information 113 shown in FIG. 4, so the relationship of the pattern information 141 is “vehicle type”.

  The concept extraction unit 55 (an example of a second extraction unit) uses the syntax tree pattern stored in the pattern information storage unit 36 to extract a new concept pair from the document data and stores it in the concept information storage unit 31. .

  Specifically, the concept extraction unit 55 reads pattern information from the extraction pattern information storage unit 37 or the user pattern information storage unit 38, and the acquisition unit 45 acquires a sentence having a syntax tree pattern of the read pattern information. Extract from document data.

  FIG. 10 is a diagram illustrating an example of the syntax tree pattern of the pattern information 117 illustrated in FIG. The syntax tree pattern shown in FIG. 10 is composed of character strings 153 to 155.

  FIG. 11 is a diagram illustrating an example of syntax tree information of a sentence having the syntax tree pattern illustrated in FIG. The syntax tree information shown in FIG. 11 is the syntax tree information of a sentence composed of character strings 161 to 165, and the character strings 163 to 165 correspond to the character strings 153 to 155 of the syntax tree pattern shown in FIG. ing. Therefore, the sentence composed of the character strings 161 to 165 shown in FIG. 11 is extracted by the concept extraction unit 55.

  Then, the concept extraction unit 55 examines the correlation of the vocabulary including the variable part in the syntax tree pattern of the extracted sentence. If there is a correlation, the concept extraction unit 55 determines the character string corresponding to the variable part of the syntax tree pattern. Extracted as a concept pair and stored in the extracted concept information storage unit 32 together with the relationship and label.

  FIG. 12 is a diagram illustrating an example of a state in which the concept information including the concept pair extracted from the sentence illustrated in FIG. 11 is newly stored in the extracted concept information storage unit 32. In the example illustrated in FIG. 12, the concept pair that is a pair of characters included in the character string 163 and the character string 164 illustrated in FIG. 11 is stored in the extracted concept information storage unit 32 as the concept information 171 together with the relationship, the label, and the contradiction. ing. Since the pattern information used for extracting the concept pair is the pattern information 117 shown in FIG. 6, the relationship of the concept information 171 is “vehicle type”. The label and contradiction will be described later.

  The generation unit 60 generates an ontology using a plurality of concept pairs stored in the concept information storage unit 31. Specifically, the generation unit 60 reads the concept information from the extracted concept information storage unit 32 or the user concept information storage unit 33, and together with the nodes assigned the names of the concept pairs of the read concept information, Create a link.

  FIG. 13 is a diagram illustrating an example of an ontology generated using the concept information 113 and 114 illustrated in FIG. 4 and the concept pair of the concept information 111 illustrated in FIG. The ontology shown in FIG. 13 is an ontology composed of nodes 181 to 183, and a link circulates between the nodes 181 and 182.

  The determination unit 65 determines whether there is a contradiction between concepts constituting the ontology generated by the generation unit 60. Specifically, the determination unit 65 determines that there is a contradiction between concepts when a cycle occurs between the concepts constituting the ontology generated by the generation unit 60. Then, the determination unit 65 causes the extracted concept information storage unit 32 or the user concept information storage unit 33 to store the presence or absence of a contradiction as a determination result in association with the concept pair used for the ontology generation.

  In the example shown in FIG. 13, since the link circulates between the node 181 and the node 182, the determination unit 65 determines that there is a contradiction in the concept pair of the generation source of the node 181 and the node 182 as shown in FIG. To do.

  FIG. 15 is a diagram illustrating an example of a state in which the presence or absence of contradiction is stored in the extracted concept information storage unit 32 by the ontology contradiction determination illustrated in FIG. 14. FIG. 16 is a diagram illustrating the ontology contradiction determination illustrated in FIG. FIG. 5 is a diagram illustrating an example of a state in which the presence or absence of contradiction is stored in a user concept information storage unit 33.

  In the ontology shown in FIG. 13, since it is determined that there is a contradiction in the concept pair of the generation source of the node 181 and the node 182, the concept information 111 in the extracted concept information storage unit 32 shown in FIG. 15 and the user concept shown in FIG. Inconsistency is set in the concept information 114 of the information storage unit 33.

  The output control unit 70 causes the output unit 20 to output the determination result of the determination unit 65 together with the ontology generated by the generation unit 60. For example, as illustrated in FIG. 14, the output control unit 70 causes the output unit 20 to highlight a portion where a contradiction occurs. Further, the output control unit 70 causes the output unit 20 to output a comparison result of the comparison unit 80 described later.

  When the registration operation is accepted by the accepting unit 40, the registration unit 75 registers the accepted concept pair in the user concept information storage unit 33 or registers the accepted syntax tree pattern in the user pattern information storage unit 38.

  When a comparison operation is received by the receiving unit 40, the comparing unit 80 compares labels for the same concept pair stored in both the extracted concept information storage unit 32 and the user concept information storage unit 33, and an extracted pattern information storage unit. 37 and label comparison for the same syntax tree pattern stored in both the user pattern information storage unit 38 and the user pattern information storage unit 38.

  Next, the operation of the ontology generation device according to the present embodiment will be described.

  FIG. 17 is a flowchart illustrating an example of a flow of an ontology generation process performed by the ontology generation apparatus 1 according to the present embodiment.

  First, when the acquisition unit 45 receives an acquisition operation, the acquisition unit 45 acquires a document data group (step S10).

  Subsequently, the pattern extraction unit 50 is syntax tree information of a sentence in which a concept pair of concept information stored in the concept information storage unit 31 co-occurs from a document acquired by the acquisition unit 45, and the concept pair Pattern information extraction processing is performed for extracting a syntax tree pattern indicating a dependency relationship between the syntax tree in which each of the concepts is replaced with a variable and another syntax tree and storing the extracted syntax tree pattern in the pattern information storage unit 36 (step S20). Details of the pattern information extraction process will be described later.

  Subsequently, the concept extraction unit 55 extracts a new concept pair from the document data acquired by the acquisition unit 45 using the syntax tree pattern stored in the pattern information storage unit 36, and stores it in the concept information storage unit 31. A concept information extraction process is performed (step S30). Details of the concept information extraction process will be described later.

  Subsequently, when the concept information including the new concept pair is stored in the concept information storage unit 31 by the concept extraction unit 55 and the number of concept information in the concept information storage unit 31 increases (Yes in step S40), the generation unit 60 performs an ontology generation process using the concept pair of concept information stored in the concept information storage unit 31 (step S50). Details of the ontology generation process will be described later.

  Subsequently, the determination unit 65 performs a contradiction determination process for determining whether there is a contradiction between concepts constituting the ontology generated by the generation unit 60 (step S60). Details of the contradiction determination process will be described later.

  Subsequently, the output control unit 70 causes the output unit 20 to output the determination result of the determination unit 65 together with the ontology generated by the generation unit 60 (step S70).

  FIG. 18 is a flowchart showing an example of the flow of the pattern information extraction process shown in step S20 of FIG.

  First, the pattern extraction unit 50 selects concept information to be read from the extracted concept information storage unit 32 or the user concept information storage unit 33 (step S202).

  At this time, the pattern extraction unit 50 stores the concept information having the same concept pair in the extracted concept information storage unit 32 and the user concept information storage unit 33, and stores the user concept information storage if the concept pairs have different labels. The concept information stored in the unit 33 is selected.

  For example, the concept information 112 stored in the extracted concept information storage unit 32 shown in FIG. 3 and the concept information 113 stored in the user concept information storage unit 33 shown in FIG. Since they are different from each other, the pattern extraction unit 50 selects the concept information 113 stored in the user concept information storage unit 33.

  Subsequently, the pattern extraction unit 50 reads the selected concept information and searches the document data acquired by the acquisition unit 45 for document data in which the concept pairs of the read concept information co-occur (step S204).

  Subsequently, the pattern extraction unit 50 parses the retrieved document data, and extracts a sentence in which concept pairs co-occur as syntax tree information (step S206). In the example shown in FIG. 7, the pattern extraction unit 50 uses the concept information 113 shown in FIG. 4 to extract sentences composed of character strings 121 to 125 as syntax tree information.

  Subsequently, the pattern extraction unit 50 deletes an unnecessary character string from the syntax tree information of the extracted sentence, and newly extracts a syntax tree pattern in which the character string corresponding to the concept information is replaced with a variable. (Step S208).

  In the example illustrated in FIG. 8, the pattern extraction unit 50 deletes the character string 121 from the syntax tree information illustrated in FIG. 7, and character strings 122 and 124 corresponding to the concept pairs in the concept information 113 are character strings including variables, respectively. 132 and 134 are replaced.

  Subsequently, the pattern extraction unit 50 performs machine learning of the syntax tree pattern using the character strings (subtrees) constituting the extracted syntax tree information, the concept pairs used for the extraction, and the labels of the concept pairs as features, thereby syntactic syntax. A classifier used for labeling the tree pattern is generated (step S210). An example of the machine learning method is SVM, but is not limited thereto.

  Here, the classifier will be described with reference to FIGS.

  FIG. 19 is a diagram illustrating an example of a vector used for machine learning of a syntax tree pattern. In the example shown in FIG. 19, the character strings 121 to 125 constituting the syntax tree information shown in FIG. 7, the concept pair of the concept information 113 shown in FIG. 4 used for extracting the syntax tree pattern shown in FIG. A vector is generated with a label indicating a positive example as a feature.

  Specifically, in the example illustrated in FIG. 19, a label indicating a positive example of the concept information 113, a character string 121 to 125, a concept pair of the concept information 113, and a dependency pair of the character string 121 to 125 and the concept information 113. A vector is generated for each of the relationships.

  FIG. 20 is a diagram illustrating an example of a concept pair and a sentence in which the concept pair co-occurs. In the example illustrated in FIG. 20, sentences 221 to 225 co-occur in the concept pairs 211 to 215, respectively.

  FIG. 21 is a diagram illustrating an example of a vector set generated from the concept pair illustrated in FIG. 20 and a sentence in which the concept pair co-occurs. In the example shown in FIG. 21, the table 227 follows the right column of the table 226, and the table 228 continues after the right column of the table 227.

  FIG. 22 is a diagram illustrating an example of the concept of a classifier generated by machine learning of the vector set shown in FIG. 21, and FIG. 23 is generated by machine learning of the vector set shown in FIG. It is a figure which shows an example of a classifier. In the example illustrated in FIG. 23, the table 242 continues to the right column of the table 241, and the table 243 continues to the right column of the table 242.

  In the example shown in FIG. 22, “positive example” and “negative example” vectors exist on the vector space 230, and are divided into a negative example vector space 231 and a positive example vector space 232 by a boundary curve 233. ing. Therefore, by inputting the vector set to the classifier shown in FIG. 22, it is possible to determine whether the label to be given to the syntax tree pattern is a positive example or a negative example.

  Returning to FIG. 18, the pattern extraction unit 50 repeats the processing from step S <b> 202 to step S <b> 210 until all selectable concept information stored in the extraction concept information storage unit 32 and the user concept information storage unit 33 is processed ( No in step S212).

  If all selectable conceptual information has been processed (Yes in step S212), the pattern extraction unit 50 selects the extracted syntax tree pattern and inputs it to the classifier (step S214).

  Subsequently, the pattern extraction unit 50 acquires a classification label associated with the input syntax tree pattern from the classifier, and stores it in the extraction pattern information storage unit 37 and the user pattern information storage unit 38 in association with the syntax tree pattern ( Step S216). For example, when the syntax tree pattern shown in FIG. 8 is input to the classifier shown in FIG. 23, since any character string (syntax tree) is a positive example, the syntax tree pattern shown in FIG. It is associated.

  Subsequently, the pattern extraction unit 50 repeats the processing from step S214 to step S216 until all the extracted syntax tree patterns are processed (No in step S218). If all the extracted syntax tree patterns have been processed (Yes in step S218), the pattern extraction unit 50 ends the process.

  FIG. 24 is a flowchart illustrating an example of a procedure flow of the conceptual information extraction process illustrated in step S30 of FIG.

  First, the concept extraction unit 55 selects pattern information to be read from the extraction pattern information storage unit 37 or the user pattern information storage unit 38 (step S302).

  At this time, the concept extraction unit 55 stores pattern information having the same syntax tree pattern in the extraction pattern information storage unit 37 and the user pattern information storage unit 38, and if the two syntax tree patterns have different labels, the user pattern The pattern information stored in the information storage unit 38 is selected.

  For example, the pattern information 116 stored in the extracted pattern information storage unit 37 shown in FIG. 5 and the pattern information 117 stored in the user pattern information storage unit 38 shown in FIG. Are different from each other, the concept extraction unit 55 selects the pattern information 117 stored in the user pattern information storage unit 38.

  Subsequently, the concept extraction unit 55 reads the selected pattern information and checks whether or not the syntax tree pattern of the read pattern information is a positive example (step S304).

  Subsequently, when the selected syntax tree pattern is a correct example (Yes in step S304), the concept extraction unit 55 selects a sentence having the syntax tree pattern from the document data acquired by the acquisition unit 45. Extract (step S306). In the example illustrated in FIG. 11, the concept extraction unit 55 extracts a sentence including character strings 161 to 165 having the syntax tree information illustrated in FIG. 10. When the selected syntax tree pattern is a negative example (No in step S304), the concept extraction unit 55 selects new pattern information.

  Subsequently, the concept extraction unit 55 extracts vocabulary pairs corresponding to variables of the syntax tree pattern as concept information (step S308). In the example illustrated in FIG. 11, the concept extraction unit 55 extracts a character pair included in the character string 163 and the character string 164 as a concept pair.

  Subsequently, the concept extraction unit 55 statistically tests the extracted concept information (step S310). Note that, for example, a chi-square test can be used for the statistical test of the concept information, but the present invention is not limited to this.

  FIG. 25 is a diagram showing an example in which the correlation between the concept pairs in the concept information 171 in FIG. 12 is tested by chi-square test. In the example shown in FIG. 25, the appearance frequency and the co-occurrence frequency of each concept pair of the concept information 171 for the document data are shown, and the co-occurrence frequency is shown when both concept pairs appear.

  And the correlation of each concept pair can be statistically determined by determining the independence regarding the appearance of each concept pair. Independence can be determined using a chi-square test, which is one of the test methods often used in statistical analysis, but is not limited to this. Here, the null hypothesis is “concept pairs are independent of each other” and the alternative hypothesis is “SUV and concept pairs are not independent”.

  In the example shown in FIG. 25, the chi-square statistic is 6.2071, and the p-value is 0.013. That is, in the chi-square distribution with one degree of freedom, the probability that the chi-square statistic 6.2071 will occur is 0.0013. Since the p-value is smaller than the significance level 0.05, the null hypothesis is rejected, and the concept pairs are not independent of each other and are considered statistically correlated.

  For this reason, in the example shown in FIG. 25, the concept extraction part 55 gives the label of a positive example with respect to a concept pair. Note that, when it is determined that they are independent from each other as a result of the chi-square test, the concept extraction unit 55 gives a negative example label to the concept pair.

  Returning to FIG. 24, when the concept pair has a correlation (Yes in step S312), the concept extraction unit 55 associates the concept pair with a label indicating a positive example, or extracts the concept information storage unit 32 or the user concept information. Store in the storage unit 33 (step S314). On the other hand, when the concept pair has no correlation (No in step S312), the concept extraction unit 55 associates the concept pair with a label indicating a negative example, or the extracted concept information storage unit 32 or the user concept information storage unit 33. (Step S316).

  Subsequently, the concept extraction unit 55 repeats the processing from step S302 to step S316 until all selectable pattern information is processed (No in step S318). When all the selectable pattern information has been processed (Yes in step S318), the concept extraction unit 55 ends the process.

  FIG. 26 is a flowchart illustrating an example of the flow of the ontology generation process shown in step S50 of FIG.

  First, the generation unit 60 selects concept information from the extracted concept information storage unit 32 or the user concept information storage unit 33 (step S502).

  Subsequently, the generation unit 60 confirms whether the concept pair of the selected concept information is a positive example (Step S503).

  Subsequently, when the selected concept pair is a positive example (Yes in step S503), the generation unit 60 creates a link between the nodes together with the node to which the name of the concept pair is assigned (step S504). ). In the example illustrated in FIG. 13, the generation unit 60 generates an ontology including nodes 181 to 183.

  Subsequently, the generation unit 60 repeats the processing from step S502 to step S504 until all the conceptual information is processed (No in step S506). If all the conceptual information has been processed (Yes in step S506), the generation unit 60 ends the process.

  FIG. 27 is a flowchart showing an example of the flow of the contradiction determination process shown in step S60 of FIG.

  First, the determination unit 65 selects a node from the ontology generated by the generation unit 60 (step S602).

  Subsequently, the determination unit 65 performs a depth-first search on the ontology using the selected node as a starting point (step S604).

  Subsequently, the determination unit 65 confirms whether or not the visited node has been reached as a result of the depth-first search (step S606).

  Subsequently, when the visited node is reached (Yes in step S606), the determination unit 65 determines that there is a contradiction in the concept pairs that are the generation sources of all visited nodes visited in the depth-first search. The associated information is stored in the extracted concept information storage unit 32 or the user concept information storage unit 33 (step S608). If the visited node is not reached during the depth-first search (No in step S606), the determination unit 65 does not perform the process in step S608. In the example illustrated in FIG. 14, since the link circulates between the node 181 and the node 182, the determination unit 65 determines that there is a contradiction in the concept pair of the generation source of the node 181 and the node 182.

  Subsequently, the determination unit 65 repeats the processes in steps S602 to S608 until all the nodes are processed (No in step S610). If all nodes have been processed (Yes in step S610), the determination unit 65 ends the process.

  FIG. 28 is a flowchart illustrating an example of a flow of a procedure for registering concept information in the user concept information storage unit 33 performed by the ontology generation device 1 according to the present embodiment. In the example shown in FIG. 28, the concept information registration process will be described, but pattern information can also be registered by the same method.

  First, the receiving unit 40 receives an input of a concept information registration operation from the input unit 10 (step S802).

  Subsequently, the registration unit 75 registers (stores) the received concept information in the user concept information storage unit 33 (step S804).

  FIG. 29 is a flowchart illustrating an example of a flow of a conceptual information comparison process performed by the ontology generation device 1 according to the present embodiment. In the example shown in FIG. 29, the concept information comparison process will be described, but pattern information can also be compared using the same method.

  First, the receiving unit 40 receives an input of a conceptual information comparison operation from the input unit 10 (step S902).

  Subsequently, when the comparison unit 80 receives the input of the comparison operation, the comparison unit 80 displays the concept information having the same concept pair stored in both the extracted concept information storage unit 32 and the user concept information storage unit 33. Read (step S904).

  Subsequently, the comparison unit 80 compares the labels of the read concept information (step S906).

  If the concept information does not match due to label mismatch (No in step S908), the output control unit 70 causes the output unit 20 to output that the concept information does not match (step S910). Note that if the conceptual information matches due to matching of labels (Yes in step S908), the output control unit 70 does not perform the process shown in step S910.

  Subsequently, the comparison unit 80 repeats the processing from step S902 to step S910 until all the same concept pairs stored in both the extracted concept information storage unit 32 and the user concept information storage unit 33 are processed (step S912). No).

  If all the same concept pairs have been processed (Yes in step S912), the comparison unit 80 ends the process.

  As described above, in this embodiment, not only the ontology is generated and output, but also whether there is a contradiction between concepts constituting the ontology is determined and output. The user can easily discriminate the contradiction.

  In this embodiment, the determination result of the existence of contradiction between the concepts constituting the ontology is stored in association with the concept constituting the ontology. An ontology that can easily determine the presence or absence can be constructed.

  In the present embodiment, the syntax tree pattern is extracted using the concept pair and the concept pair is extracted using the extracted syntax tree pattern repeatedly. You can earn a wide range.

  Further, in the present embodiment, since labels are associated with concept pairs and syntax tree patterns, concept pairs that are not in a correct relationship and syntax tree patterns that are not suitable for extracting concept pairs can be excluded.

  In the present embodiment, not only the concept pair and the syntax tree pattern can be extracted by the ontology generation device itself, but also registration by the user can be used together.

  The ontology generation device 1 according to the present embodiment includes a control device such as a CPU (Central Processing Unit), a storage device such as a ROM (Read Only Memory) and a RAM (Random Access Memory), a display device such as a liquid crystal display, a keyboard, The hardware configuration includes an input device such as a mouse, a communication I / F that communicates by connecting to a network.

  The present invention is not limited to the above-described embodiment as it is, and can be embodied by modifying the constituent elements without departing from the scope of the invention in the implementation stage. In addition, various inventions can be formed by appropriately combining a plurality of constituent elements disclosed in the above embodiments. For example, some components may be deleted from all the components shown in the embodiment. Furthermore, constituent elements over different embodiments may be appropriately combined.

DESCRIPTION OF SYMBOLS 1 Ontology production | generation apparatus 10 Input part 20 Output part 30 Storage part 31 Concept information storage part 32 Extraction concept information storage part 33 User concept information storage part 36 Pattern information storage part 37 Extraction pattern information storage part 38 User pattern information storage part 40 Reception part 45 acquisition unit 50 pattern extraction unit 55 concept extraction unit 60 generation unit 65 determination unit 70 output control unit 75 registration unit 80 comparison unit 111, 112, 113, 114, 171 concept information 116, 117, 141 pattern information 121-125, 132 , 134, 153 to 155, 161 to 165 Character string 181 to 183 Node 211 to 215 Concept pair 221 to 225 Sentence 230 Vector space 231 Negative example vector space 232 Positive example vector space 233 Boundary curve

Claims (8)

An acquisition unit for acquiring document data;
A concept information storage unit for storing a concept pair which is a set of vocabulary concepts;
From the document data, among the character strings of the sentences in which the concept pair co-occurs, the dependency relationship between the first and second character strings obtained by replacing each vocabulary representing the concept of the concept pair with a variable and other character strings A first extraction unit for extracting a pattern to be shown;
A pattern information storage unit for storing the extracted pattern;
A second extraction unit that extracts a new concept pair from the document data using the pattern stored in the pattern information storage unit and stores the new concept pair in the concept information storage unit;
A generation unit that generates an ontology that organizes relationships between concepts using the plurality of concept pairs stored in the concept information storage unit;
A determination unit for determining the presence or absence of contradiction between concepts constituting the ontology;
And an output unit that outputs a determination result together with the ontology.   The ontology generation apparatus according to claim 1, wherein the determination unit determines that there is a contradiction between the concepts when a cycle occurs between the concepts constituting the ontology.   The ontology generation apparatus according to claim 1, wherein the determination unit stores the determination result in the concept information storage unit in association with the concept pair. The concept information storage unit further stores a first label indicating whether the relationship of the concept pair is correct or not in association with the concept pair,
The pattern information storage unit further stores a second label indicating whether or not the pattern is effective for extraction of the new concept pair in association with the pattern,
The first extraction unit performs machine learning using the linguistic information constituting the extracted pattern, the concept pair used for extraction, and the first label of the concept pair as a feature, thereby the first of the extracted pattern. 2 labels are generated, stored in the pattern information storage unit in association with the extracted pattern,
The second extraction unit extracts the new concept pair from the document data using the pattern in which the second label is valid among the patterns stored in the pattern information storage unit, and Store it in the concept information storage unit,
The said generation part produces | generates the said ontology using the said concept pair in which the said 1st label shows validity among the several said concept pairs memorize | stored in the said concept information storage part. The ontology generation device according to any one of? The concept information storage unit includes a first concept information storage unit that stores the new concept pair extracted by the second extraction unit, and a second concept information storage unit that stores the concept pair registered by a user. Including,
The pattern information storage unit includes a first pattern information storage unit that stores the pattern extracted by the first extraction unit, and a second pattern information storage unit that stores the pattern registered by a user,
The first extraction unit extracts the pattern from the document data by using the concept pair stored in the first concept information storage unit and the second concept information storage unit, and the first concept Store it in the information storage unit,
The second extraction unit extracts the new concept pair from the document data using the patterns stored in the first pattern information storage unit and the second pattern information storage unit, and The ontology generation apparatus according to claim 4, wherein the ontology generation apparatus is stored in one concept information storage unit. An accepting unit that accepts input of at least one of the concept pair and the pattern;
The registration unit that registers the accepted concept pair in the second concept information storage unit and registers the accepted pattern in the second pattern information storage unit. The ontology generation device described. Comparison of the first label with respect to the same concept pair stored in both the first concept information storage unit and the second concept information storage unit, and the first pattern information storage unit and the second pattern information storage unit A comparison unit that performs at least one of the comparison of the second label with respect to the same pattern stored in both of
The ontology generation apparatus according to claim 5, wherein the output unit further outputs a comparison result. An acquisition step in which the acquisition unit acquires document data;
The first extraction unit stores, from the document data, a character string of the concept pair out of a character string of a sentence co-occurring with the concept pair stored in a concept information storage unit that stores a concept pair that is a set of vocabulary concepts. A first extraction step of extracting a pattern indicating a dependency relationship between the first and second character strings obtained by replacing each vocabulary representing a concept with a variable and another character string and storing the pattern in a pattern information storage unit;
A second extraction step in which a second extraction unit extracts a new concept pair from the document data using the pattern stored in the pattern information storage unit and stores the new concept pair in the concept information storage unit;
A generation step of generating an ontology in which relationships between concepts are organized using the plurality of concept pairs stored in the concept information storage unit;
A determination step for determining whether or not there is a contradiction between concepts constituting the ontology; and
And an output control step for causing the output unit to output a determination result together with the ontology.

Images