KR101086996B1 - Apparatus for generating ontology and method thereof - Google Patents

Abstract

The present invention relates to an ontology generating device and a method thereof, in order to minimize the time of human intervention and to minimize the ontology construction time and to continuously expand the ontology, syntactic-based association information (eg, related words) An ontology generating device and method for generating semantic based ontology through matching between a graph) and a semantic-based ontology-like system (eg, ontology) are provided.

To this end, the present invention, in the ontology generation method, language-based matching step of matching between the term of the association graph and the term of the ontology by reflecting the characteristics of the language resources; A contextual information-based matching step of matching the matched relationship in the language-based matching step based on contextual information; And a weighting step of assigning a semantic weight to a link relationship between the association graph and the ontology based on the matched relationship in the contextual information-based matching step.

Ontology Generation, Ontology Matching, Association Graph, Syntax Relationship, Association Information, Ontology Similarity System, Semantic Relationship, Semantic Based Ontology

Description

Ontology generator and its method {APPARATUS FOR GENERATING ONTOLOGY AND METHOD THEREOF}

The present invention relates to an ontology generating apparatus and a method thereof, and more particularly, to ontology-like ontology including association information (e.g. association graph) representing a syntactic relationship between terms existing on a web document. The present invention relates to an ontology generating apparatus and a method for generating a semantic based ontology by matching a system (indicating a meaningful relationship).

In the following embodiment of the present invention will be described in the case of a term included in a general web document as an example, but the present invention is not limited thereto.

In the present invention, an ontology-like system refers to an ontology, a dictionary, a thesaurus, a taxanomi, or a folksonomi.

In the following embodiments of the present invention, a related graph is described as an example of related information, and an ontology is described as an example of an ontology-like system, but the present invention is not limited thereto. Put it.

In general, ontology is used to support common semantic understanding and interoperability between different meanings. Ontology matching technology is a technology that must be involved in generating an ontology because the semantic web has a heterogeneous distributed environment like the web. The ontology itself generated in this way also contains heterogeneous environments. Therefore, the ontology matching scheme is proceeding as a way of discovering semantic relationships between different ontologies. However, until now, the ontology construction process is very difficult and ontologies are constructed depending on experts in each field, and only a matching method between a few ontologies has been proposed.

Accordingly, even though the ontology is a core component of the semantic web, it has not been able to build a large scale ontology to provide a commercial service.

This will be described in more detail with reference to FIG. 1 as follows.

1 is a diagram illustrating an example of a conventional matching method between ontologies.

Referring to FIG. 1, a conventional matching method between ontologies is a matching method performed under the assumption that an object of matching is an ontology.

The matching method between ontologies is matched between ontologies by using language resources, stochastic reasoning, and integration of various approaches depending on factors such as domain identity, possibility of structural change, and creation of new ontology. Is the way to do it.

However, relations composed of associative graphs do not have the characteristics of ontologies, and even if they have a structure similar to ontologies, they cannot be used as a matching technique between ontologies.

Therefore, if a large number of web documents are produced in various ways and can be used on the semantic web, one huge ontology is matched with an ontology that can be understood by both human and machine at the same time. There is an urgent need for a way to generate a solution.

As described above, in the conventional technology, only matching techniques between ontologies have been proposed. Accordingly, the ontology is realistically constructed due to excessive time and manpower consumption for developing various applications of the semantic web by constructing a large scale ontology. There is a problem that it is difficult to expand the information continuously generated in the ontology, and it is a problem of the present invention to solve this problem and to meet the needs.

Therefore, in order to minimize ontology construction time and to continuously extend ontology while minimizing human intervention time, the present invention is syntactic-based association information (eg association graph) and semantic-based. It is an object of the present invention to provide an ontology generating apparatus and method for generating a semantic based ontology through matching between ontologie-like systems (eg, ontology).

In addition, the present invention clusters an association graph including a syntactic relationship between terms existing in a web document, and then ontology-like system, language-based matching, and context-based matching, including ontology. To provide an ontology generating apparatus and method for generating a semantic based ontology by matching association graphs and ontology similar systems through weighting, matching suitability checking, and missing relationship matching process. There is this.

The objects of the present invention are not limited to the above-mentioned objects, and other objects and advantages of the present invention which are not mentioned can be understood by the following description, and will be more clearly understood by the embodiments of the present invention. Also, it will be readily appreciated that the objects and advantages of the present invention may be realized by the means and combinations thereof indicated in the claims.

In accordance with another aspect of the present invention, an apparatus for generating an ontology includes: a language-based matching unit for performing language-based matching between association information and an ontology-like system; A contextual information-based matching unit for matching the relationship matched by the language-based matching unit based on contextual information; And a weighting unit for assigning a semantic weight between the association information and the ontology-like system based on the relationship matched in the contextual information-based matching unit.

The apparatus may further include a clustering unit configured to receive pre-established association information and an ontology-like system, cluster the association information, and transmit the clustered association information to the language-based matching unit; A checker for receiving a matching relationship from the weighting unit, and checking whether a term consistently matches an ontology-like system in a clustering unit of the clustering unit; And a forced matching unit for modifying the ontology-like system from the inspection unit or forcibly matching a missing term to the ontology-like system according to information from an external expert.

On the other hand, the method of the present invention for achieving the above object, Ontology generation method, Language-based matching step for matching between the terms of the association graph and the terms of the ontology reflecting the characteristics of the language resources; A contextual information-based matching step of matching the matched relationship in the language-based matching step based on contextual information; And a weighting step of assigning a semantic weight to a link relationship between the association graph and the ontology based on the matched relationship in the contextual information-based matching step.

In addition, the method of the present invention, the clustering step of receiving the pre-established association word graph and the ontology clustering the association word graph to the language-based matching step; Receiving a matching relationship from the weighting step and checking whether a term is consistently matched to an ontology in a clustering unit of the clustering step; And a forced matching step of correcting the ontology from the inspection step or forcibly matching the missing term to the ontology according to information from an external expert.

delete

delete

As such, the present invention clusters the association graph in order to match an association graph, which is a syntactic relationship between a number of terms existing on a web document, and an ontology-like system including an ontology, and uses a 1: N relation as a language-based matching technique. After forming, we use the contextual information-based matching technique to semantically decompose the 1: N relationship to form a 1: 1 relationship, and weight the relationship between the newly formed association word graph and the ontology.

That is, according to the present invention, a large-scale semantic basis is obtained by matching a syntax-based association graph generated by using syntactic relations (term associations) among key terms in a general web document to an ontology-like system in which semantic relations are defined. New ontology can be created.

As described above, the present invention provides a large-scale semantic-based ontology through matching between syntactic-based association information (eg, association graph) and semantic-based ontology-like system (eg, ontology). By generating the, it is possible to minimize the ontology construction time while minimizing the human intervention time, there is an effect that can continuously expand the ontology.

That is, the present invention is generated by using the syntactic relations (terms between terms) of existing web documents for automatic semantic generation for ontology construction automation, minimization of construction manpower, reduction of construction time, and continuous extension of ontology. A large-scale semantic-based ontology can be newly generated by matching a syntax-based association graph with an ontology-like system in which semantic relations are defined. A semantic web application service can be provided using the generated semantic-based ontology.

In addition, the present invention is the expenditure of the search and management of the web document including comments and tags generated innumerably in accordance with the increase of search activity for the web document and the increase of images and multimedia content, which is increasing exponentially in recent years. The user's input time, cognitive burden, and psychological burden can be minimized, and the efficiency and convenience of search can be maximized, and the meaning of content (text, image, multimedia, etc.) By leveraging semantic web technology to understand and process, there is an excellent effect to easily build an ontology that both humans and machines (software ontology) can understand.

In addition, the ontology generated in accordance with the present invention is suitable for application to semantic web applications, in particular semantic search, semantic data integration, and the like.

BRIEF DESCRIPTION OF THE DRAWINGS The above and other objects, features and advantages of the present invention will become more apparent from the following detailed description of the present invention when taken in conjunction with the accompanying drawings, It can be easily carried out. In the following description, well-known functions or constructions are not described in detail since they would obscure the invention in unnecessary detail.

First, the semantic web technology will be described to help understand the present invention.

The World Wide Web (WWW), proposed by Tim Berners-Lee in 1989, expresses information in a simple markup language called HyperText Markup Language (HTML) to share information in the Internet space. By doing so, it has revolutionized Internet-based information sharing and delivery. As a result, a huge information space was established that interconnected various information resources, and the Internet rapidly spread to the real world, thereby innovating the information society.

However, the existing web technologies face many problems as the amount of web information increases. In particular, the existing web only permits access to information by keywords, and according to the simple keyword search, a myriad of unnecessary information has been mass produced, resulting in a flood of information. In addition, the existing web can be said to be a human-centered information processing technology in which a computer does not provide enough functions to effectively extract, interpret, and process necessary information on its own.

Therefore, the semantic web is a technology that can extend the existing web to realize semantic interoperability based on well-defined meanings that computers can understand and to build an effective cooperative system between humans and computers. It appeared.

The Semantic Web is a semantic-based search that allows not only people but machines (computers) to understand the information that exists on the web, and finds only the results that meet the user's needs, and facilitates collaboration between people and machines or machines and machines. By doing so, it refers to a web that can be used automatically for people.

In other words, the semantic web is a next-generation web technology that enables a computer to understand, automate, integrate, and reuse the meaning of information resources.

1) Ontology

Ontology is a formal specification system for shared conceptualization and provides semantic information of domain vocabulary. Ontology is a kind of knowledge expression, and the computer can understand the concept represented by the ontology and process the knowledge. In order to deal with inferences, the ontology's axiom and rule system are needed.

2) semantically annotated web

A semantically annotated web is an ontology annotated web, which is a knowledge base. The Semantic Web can build a huge knowledge base that semantically integrates the distributed information resources of the Internet. In a narrow sense, it may be possible to build a knowledge base of information resources of a company or institution.

3) agent

An agent is an intelligent agent that collects, retrieves and infers information resources on behalf of a person (user), and exchanges information with other agents. Intelligent agents are the core of semantic web-based application systems.

The semantic web realizes semantic interoperability by using ontology and agent technology, and thus, the semantic web can leap from the information-based web to the knowledge-based semantic web.

Next, in order to help the understanding of the present invention, comparing the matching method between ontologies, which are conventional methods, and the matching method between the associated word graph and the ontology (ontology similar system) according to the present invention, there are differences as follows.

First, each ontology used in a matching method between ontologies, which is a conventional method, is a semantic relationship, but an associative graph used in the present invention is a syntactic relationship. Therefore, the existing matching method between ontology is a matching problem between the semantic relationship diagram, but the matching method between the associative graph and the ontology according to the present invention is a matching problem between the syntactic relationship and the semantic relationship diagram.

Second, the individual ontology targeted by the matching method between the ontology includes the characteristics of the ontology (e.g., concept, attribute, relationship, constraint, axiom, instance, etc.) The association word graph in the ontology matching method does not have the characteristics of the ontology because it is a relational diagram constructed based on the degree of simultaneous occurrence between words. Therefore, the matching method between the association word graph and the ontology according to the present invention is a matching problem between the ontology and the non-ontology.

Third, in the conventional method of matching between ontologies, all target ontology forms a perfect semantic relationship (eg, parent-child relationship, sibling relationship, etc.), but in the method of matching between graph and ontology according to the present invention Since these are syntactic relations, the result after matching only forms a semantic relation only partially, and remains partially syntactic.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings.

2 is a configuration diagram of an ontology generating system to which the present invention is applied and an ontology generating device according to a preferred embodiment of the present invention.

First, referring to FIG. 2, an ontology generation system to which the present invention is applied will be described as follows.

As shown in FIG. 2, the ontology generation system to which the present invention is applied may generate air (Co-occurrence Frequency) using the ontology 10 and various documents existing on the web based on the knowledge of domain experts. The ontology generation device 30 for generating a semantic-based ontology by matching the association word graph with an ontology (ontology similar system) based on the pre-established association word graph 20 is generated by the ontology generation device 30. Semantic web application application for the user terminal or semantic web application to provide a variety of semantic web application services using the ontology storage 40 for storing the ontology, and the ontology stored in the ontology storage 40 (Eg, software agent) 50.

Next, as shown in Figure 2, the ontology generation apparatus 30 according to the present invention, the language-based matching unit for matching the on-ology 10 between the associated language graph 20 and the ontology 10 received from the outside (32), based on the contextual information-based matching unit 33, and the contextual information-based matching unit 33 for matching the relationship matched in the language-based matching unit 32 based on the contextual information It includes a semantic weighting unit 34 for giving a semantic weight between the association graph and the ontology.

In addition, the ontology generating apparatus 30 according to the present invention receives an associated ontology 10 and a related word graph 20, and associates the related word graph by clustering the related word graph to the language-based matching unit 32. The graph clustering unit 31 and the semantic weighting unit 34 receive the matching relationship with the semantic weights, and check whether the term is consistently matched to the ontology in the clustering unit of the association graph. Force matching unit 36 for modifying the ontology from the inspection unit 35 or forcibly matching certain missing terms to the established ontology in accordance with the inspection unit 35 and correction or supplemental information from an external domain expert. More).

Here, referring to FIG. 2, the components are described in more detail as follows.

First, the associated word graph clustering unit 31 receives the pre-established ontology 10 and the associated word graph 20 to determine whether the related term target terms are matched to the ontology in a predetermined distribution in a subsequent process. As a preliminary work to make this possible, cluster the association graph.

The language-based matching unit 32 reflects the characteristics of language resources to match the terms (terms) of the ontology graph with terms (concepts, attributes, relationships, instances, etc.) of the ontology.

The contextual information-based matching unit 33 matches the relationship of 1: N (N is a natural number) generated by the language-based matching unit 32 based on contextual information to newly generate a 1: 1 relationship. Explode the graph.

The semantic weighting unit 34 assigns a semantic weight to a link relationship between the association word graph and the ontology based on the relationship decomposed by the contextual information-based matching unit 33.

In addition, the inspection unit 35 maintains constant terms of the associated graph due to a match between a statistically generated correlation graph (eg, 3 million to 4 million words in Korean) and an ontology term (hundreds of thousands in the Korean language). The semantic weighting unit 34 receives the matching relationship with the semantic weighting so as to determine whether the ontology is matched, and clusters the association word graph clustering unit in the association word clustering unit 31. Check that some of the terms contained in are consistently matched to the ontology.

The forced matching unit 36 may determine that the terms in the associative graph clustering unit are not matched with the ontology, even if only one term is not matched with the ontology. Certain terms in the table are forcibly matched to an existing ontology.

Next, each component of the ontology generating device and its operation method according to the present invention will be described in detail as follows.

First, in matching between the association information and the ontology-like system, the ontology 10 to be matched with the association word graph 20 is collected or secured from the outside.

At this time, the ontology-like system uses not only the ontology 10 pre-generated by domain experts, but also thesaurus, texanomi, or folksonomi used in the classification of dictionaries, vocabularies, etc. that express the meaning of the existing vocabulary. Include as a target of.

Next, referring to the method of generating the related word graph 20, first, key terms that may be the target of the related word are extracted from the web document, and in this case, stopwords / duplicate words are excluded. In this case, to extract the core term for the target correlator, the core term for the correlated term is selected using a morpheme analyzer or an ngram method. In addition, MI (Mutual Information), TF (Term Frequency) / IDF (Inversed Document Frequency: the number of documents in which a word appeared in the entire document to measure the degree of association between the extracted related words Reciprocal), C-value / NC-value, etc., the method that can measure the mutual information best. The measured correlation should satisfy the threshold condition (e.g. 40% to 50% of mutual information, or the number of connected nodes within one node, etc.) and normalization execution. Using the degree of association, an association word graph for each domain is generated by applying the graph theory with a node-arc. The domain-related association graphs generated as above are merged into one large graph through co-occurrence between nodes and generation of meaningful node relationships, thereby generating a syntax-based association graph.

On the other hand, the ontology generation device 30 includes a process of matching the ontology 10 to be matched with the graph 20 in association.

First, the association word cluster clustering unit 31 clusters the association word graph to be matched. The reason for clustering an association graph first is that there is a difference in the number between ontology terms (concepts, attributes, relationships, instances, etc.) that are subject to matching and association terms (eg, target terms of linkage between terms based on mutual information). To solve it. For example, when the ratio of the number of terms in the ontology and the number of terms in the association graph is 1: 100 or 1: 1000, clustering is performed to determine whether the terms of the ontology are matched to the association graph at a constant ratio. . The terms, which are related graph elements, contain a variety of terms compared to ontologies created by experts and include terms such as new words, compound words, and trendy words, or are used only for a certain period of time. A mismatch can occur. Therefore, if there is a part (area) consisting only of the latest terms in the association diagram of the association word graph, and if this area is clustered, it is determined whether these areas match the ontology as the clustering unit of the association word and force the corresponding area to the ontology. Cluster the associative graph to match (artificially). For the association word graph, clustering may be performed by using a method such as HCS (Highly Connected Subgraph) or Camon.

Next, the language-based matching unit 32 matches the terms of the association graph with the terms of the ontology only from a linguistic point of view. At this time, the language-based matching unit 32 performs a string-based matching process and a linguistic resource-based matching process. First, in the string-based matching process, it is assumed that similar terms use similar names or expressions. For example, a character string between an ontology concept, an expression in an instance, and the like and a target word of a related word in a related word graph is compared. In this case, the case of correct matching is identified by a 1: N relationship. In the linguistic resource-based matching process, linguistic resources are used to understand the meaning of terms and the relationship between terms. With reference to linguistic relational terms (synonyms, synonyms, abbreviations, etc.), the matching relationship between ontology concepts, instance expressions and terms that are subjects of association terms is identified as 1: N. For example, the term “Kim Hee Sun” among the related term target terms may be matched with a 1: N convention such as singer Kim Hee Sun, talent Kim Hee Sun, basketball player Kim Hee Sun, and politician Kim Hee Sun. Of course, in this case, the Chinese word list may be secured to limit the Chinese word list to match. The example will be described later in detail through a matching example between the language-based association word graph and the ontology of FIG. 3.

Next, the contextual information-based matching unit 33 matches the terminology of the associated graph with the terminology of the ontology based on the contextual information. At this time, the contextual information-based matching unit 33 performs a semantic-based matching process and a constraint-based matching process. First, the semantic-based matching process is described. In this process, different contextual information is represented by different structures, properties, and relationships, and thus associated with the concepts and properties of the ontology. Matching is performed using the primary and secondary related terms of the graph as contextual information. First of all, the ontology's context information includes super class concept, sibling concept, target property, object property, and instance. Target relevant information. On the other hand, as contextual information of the association graph, one of the primary association words of the target term of the association graph and the association words of the selected primary association term are used as situation information. When matching the association graph and the ontology, the matching having the highest similarity is selected, and the remaining matched links are released to perform matching in a 1: 1 relationship. In this process, a die coefficient function such as Equation 1 below may be used to select the highest similarity matching. For example, the union of the situation information (E) of a specific term of the ontology and the situation information (E ') corresponding to the specific term of the graph is used as the denominator, and the context information (e) of the specific term of the ontology is associated with the denominator. Only the most similar matching relationship is selected based on the intersection of the situation information E 'corresponding to a specific term of the graph as a molecule. The above example will be described in detail later through an example (1) of matching between the contextual information-based association word graph and the ontology of FIG. 4.

As a next step of the semantic-based matching process, a constraint-based matching process is performed. In this process, since the different contextual information is represented by different structures, attributes, and relationships, the term x and the ontology attribute y coincide in terms of the first and second order of the term x. If the related term (the context information of term (x)) matches the situation information of the ontology attribute (y) (the domain and range of the attribute (y)), the term (x) and the ontology attribute (y) are almost the same. Matches in relation to the object considered. The example will be described later in detail through an example (2) of matching between the contextual information-based association word graph and the ontology of FIG. 5.

Next, the semantic weighting unit 34 assigns semantic weights to the relationships between terms matched in a 1: 1 relationship through the contextual information-based matching process.

In more detail, each link decomposed into a 1: 1 relationship through the contextual information-based matching process is a newly created relationship / arc, which considers the ontology as a graph and weights the existing call. There is a need to assign a connection weight to the corresponding (correspondent). Here, the given connection weight may be applied to various graph search algorithms. In this case, a weighting value considering the reliability of the connection is required. The basic call weighting criteria assigns values between [0, 1], but gives higher values than superclasses and lower values than properties. Because ontology includes concepts, properties, and instances constructed through expert domain knowledge, and superclasses and properties are subclasses and instances. And terminology rather than terminology (Term). Therefore, we assign semantic weights in the order of upper class <Link <attribute <subclass. The detailed connection weighting method and function will be described later in detail in the semantic weighting process of FIG. 6.

Next, the inspector 35 is one of the units clustered in the association graph clustering process. Even if the term is matched to the ontology, it is determined. In other words, it checks for adequacy of matching. For example, terminology in the association graph clusters consisting of association words associated with new words may not be included in an ontology built on the knowledge of domain experts. Identify the cluster.

Next, the forced matching unit 36 forcibly connects the representative words of the clustered terms in the association graph that are not matched with the ontology in the inspection unit 35 to the ontology. This is because the associative graph is a syntactic graph and the ontology is a semantic graph. In order to construct a large-scale ontology through mutual matching, the terms of the associative graph must be matched to the terms of the ontology with a constant distribution.

Next, the huge ontology constructed through the ontology generating device 30 is stored in the ontology repository 40 and can be used by the user terminal or semantic application 50.

FIG. 3 is a diagram illustrating a matching process between a language-based association word graph and an ontology in the language-based matching unit of FIG. 2 according to an exemplary embodiment of the present invention.

Referring to FIG. 3, a large-scale ontology is constructed by matching portions separated by the ontology area 301 and the graph area 302. In the case of the term 'Kim Hee-sun' in the related word graph area 302, it is matched and linked with the relationship between 'Talent Kim Hee-sun', 'Basketball player Kim Hee-sun' and 'Political Kim Hee-sun' in the ontology area 301 in a 1: N relationship. . In addition, 'basketball player' in the association word graph area 302 is matched with 'basketball player' which is a subclass of a professional player class of the ontology area 301, and 'central zone' in the association word graph area 302 is ontology. It matches the 'major' and 'central university' synonyms of 'central university', which is an instance of school in the area 301.

In particular, in the case of 'Kim Hee-sun' matched in a 1: N relationship, it is necessary to separate which class instance of the ontology area 301 corresponds to. However, in the language-based matching process of the language-based matching unit 32, the terminology of the association graph and the term of ontology go through a string-based matching process and a linguistic resource-based matching process as described above. When matched correctly or syntactically (synonyms, synonyms, abbreviations, etc.) with semantic similarity are linked in a 1: N relationship, and the matching result is divided into '0' or '1'.

4 is a first view illustrating a matching process between a contextual information-based association word graph and an ontology in the contextual information-based matching unit of FIG. 2 according to an exemplary embodiment of the present invention.

Referring to FIG. 4, the ontology region 401 and the association word graph area 402 are separated, but as mentioned in the description of FIG. 3, the term and ontology of the association word graph in the language-based matching unit 32 are language-based. We want to separate the results matched between terms based on the situation information. Because when we try to find the value of a specific term (concept, property, instance, etc.) to be matched with the ontology, the association graph is matched with several, but the term in the association graph is a relation with the specific term of the ontology. Because it does not mean a relationship. For example, the term 'Kim Hee-sun' of the related word graph of FIG. 4 is a language-based match between the related word graph term and the ontology term. It can be semanticly separated through the status information of the graph. Ontology contextual information utilizes the superclass concept of the target term, sibling concept, relationship (Property, ObjectProperty), and superordinate concept of the instance, and the contextual information of the association graph is one of the primary related words. The association words of the selected primary association word can be used. In FIG. 4, the term 'Kim Hee-sun' of the related word graph region 402 may use 'Hyo-ri', which is a primarily selected related word, and 'singer', 'marriage', etc., which are related words of the selected primary related word. On the other hand, context information of 'Talent Kim Hee-sun', which is one of the terms of ontology area 401, is 'Talent' which is an instance concept of 'Kim Hee-sun', 'Marriage' which is a property of talent, and 'Singer' which is a concept of marriage and talent. ',' Celebrity ', a higher concept of talent, can be used as context information of ontology. As a result, the term 'Kim Hee-sun', which is a term of the associative graph area 402, is connected to the talent 'Kim Hee-sun' of an ontology, which has a similar meaning, and in the case of another 'Kim Hee-sun' which is a term of the associative graph area 402. Matching may be performed in order of high similarity. In the case of other 'Kim Hee-sun' in the related word graph region 402, it is possible to match in the order of high similarity while repeating the case of Kim Hee-sun-Dong-pyeon and Kim Hee-sun-politician-Dang-Soo of the ontology. As for the matching method, as described above in the contextual information-based matching unit 33 of FIG. 2, a dice coefficient function such as Equation 1 may be used.

FIG. 5 is a second view illustrating a matching process between a contextual information-based association word graph and an ontology in the contextual information-based matching unit of FIG. 2 according to an exemplary embodiment of the present invention.

Referring to FIG. 5, the ontology area 501 and the associative graph area 502 are separated, but as mentioned in the description of FIG. 4, based on the result of matching the related term graph term and the ontology term based on contextual information. Separate each term in semantic detail, and then further check that each linked relationship is semantically close. That is, in the case of matching through the above-described process in FIG. 4, it is a method of determining whether the terms of the associated graphs that are the targets of each link and the terms of the ontology are substantially matched. For example, the term 'singer' in the association graph and 'singer' in terms of the ontology have a primary association of 'marriage' in the singer context information in the association graph, and the concept of 'singer', the term in ontology. Since there is 'marriage' as a property of, the two terms can be judged to be almost identical. Detailed description thereof has been described above with the contextual information-based matching unit 33 of FIG. 2.

6 is a diagram illustrating a semantic weighting process in the semantic weighting unit of FIG. 2 according to an exemplary embodiment of the present invention.

Referring to FIG. 6, the ontology and associated graphs are matched in a 1: 1 relationship through the process described above with reference to FIG. 5, and in order to be used as a huge ontology, a certain semantic weight is given to the connection part as one homogeneous ontology. Must be created. Basically, the semantic weighting method gives the link weight lower than the property weight of the ontology (Superclass <Link <Property <Subclass). In addition, link weights are classified based on the constraint-based (see description of the contextual information matching process of FIG. 2) check, which is one of contextual information-based matching techniques, and the ontology is checked for a valid value. It gives a value corresponding to the attribute weight of. Even in this case, it is lower than the attribute maximum weight value of the ontology and lower than the weight value of the attribute with the document. If the attribute value of the ontology does not exist, the weight is assigned as in the case of 2) below. In particular, in connection case 2, the correction factor value should be added. Here, the correction coefficient value may have a value between 0 and 1, and preferably has a value of 0.7.

The link weighting function used in the semantic weighting process of FIG. 6 will be described in more detail as follows. In the link weighting function below, L _w represents a weight of a link, P _w represents a weight of a property, and P _d represents a weight of a property having a document. .

1) When more than one ontology attribute exists

-Link case 1: Constraint-based check result of the situation information is valid: The link weight is given higher than the attribute minimum weight as shown below, and the smaller of the attribute maximum weight and the minimum weight of the attribute with the document as follows. Give smaller than

. [min] P _w <L _w <[ min] ( [max] P _{w ,} [ min] P _d )

   -Link Case 2: If the constraint-based check result of the situation information is invalid: The link weight is given higher than the upper class weight and smaller than the attribute minimum weight as shown below.

. Parent <L _w <[ min] P _w

2) Ontology attribute does not exist

-Link Case 1: When the constraint-based check result of the situation information is valid: The link weight is given higher than the upper class weight and smaller than the lower class weight.

. Upper <L _w <lower

-Link Case 2: If the constraint-based check result of the situation information is invalid: The link multiplied by the correction factor is larger than the upper class weight and smaller than the lower class weight. link) assigns a weight.

. Top <L _w * Correction factor <lower

Here, the threshold value used as the correction coefficient has a value between 0 and 1, and preferably has a value of 0.7.

7 is a flowchart illustrating an ontology generation method according to an embodiment of the present invention.

At this time, since the specific embodiment of the ontology generation method of the present invention is as described above, only the operation of the ontology generation method will be briefly described.

First, the associated word graph clustering unit 31 receives the associated ontology 10 and the associated word graph 20 to cluster the associated word graph (701). That is, the associated word graph clustering unit 31 receives the pre-established ontology 10 and the associated word graph 20 from the outside, and in the subsequent inspection process 705, the related term target terms are defined in a constant distribution. As a preliminary work to determine whether the ontology is matched, the association graph is clustered.

Thereafter, the language-based matching unit 32 matches the association graph 20 and the ontology 10 based on language (702). That is, the language-based matching unit 32 matches the terms (terms, concepts, attributes, relationships, instances, etc.) of the ontology (ontology similar system) and the terms (terms) of the association graph by reflecting the characteristics of the language resources. .

Thereafter, the contextual information-based matching unit 33 matches the match matched by the language-based matching unit 32 based on the contextual information (703). That is, the contextual information-based matching unit 33 matches the relationship of 1: N (N is a natural number) generated by the language-based matching unit 32 based on the contextual information to newly generate a 1: 1 relationship. Explode the graph.

Subsequently, the semantic weighting unit 34 assigns semantic weights between the association word and the ontology based on the relationship matched by the contextual information-based matching unit 33 (704). That is, the semantic weighting unit 34 assigns semantic weights to the link relationship between the association word graph and the ontology based on the relationship decomposed by the contextual information-based matching unit 33.

Subsequently, the inspection unit 35 receives the matching relationship given the semantic weight from the semantic weighting unit 34 and checks whether the term is consistently matched to the ontology in the clustering unit of the association graph (705). ). That is, the terms of the associated word graph are constant due to the matching between the related word graph (eg, 3 million to 4 million words in Korean) and the ontology term (hundreds of thousands in Korean). The semantic weighting unit 34 receives the matching relationship with the semantic weights so as to determine whether the ontology is matched with the ontology, and in the associative graph graph clustering unit in the associative graph graph clustering unit 31. Check that some of the terms in the cluster consistently match the ontology.

Subsequently, the forced matching unit 36 modifies the ontology from the inspection unit 35 according to the correction or supplementation information from the external domain expert or forcibly matches the missing certain term to the already built ontology (706). That is, when the forced matching unit 36 does not match any of the terms in the associative graph clustering unit with the ontology, the ontology is modified or missing associative graph according to the correction or supplementary information from an external domain expert. Certain terms in the cluster are forcibly matched to the established ontology.

As described above, in the present invention, due to the matching between the association word graph and the ontology, it is expected that the semantic web application will add value in real life, rather than the existing method that requires a lot of human intervention and construction time.

On the other hand, the method of the present invention as described above can be written in a computer program. And the code and code segments constituting the program can be easily inferred by a computer programmer in the art. In addition, the written program is stored in a computer-readable recording medium (information storage medium), and read and executed by a computer to implement the method of the present invention. The recording medium may include any type of computer readable recording medium.

The present invention described above is capable of various substitutions, modifications, and changes without departing from the technical spirit of the present invention for those skilled in the art to which the present invention pertains. It is not limited by the drawings.

The present invention relates to an ontology generation method through matching between an association graph and an ontology, and facilitates the construction of an ontology that is indispensable in the semantic web, and thus can be practically applied to various applications of the semantic web, for example, semantic search and semantic data integration. have.

1 is a diagram illustrating an exemplary method of matching between conventional ontology;

2 is a configuration diagram of an ontology generating system to which the present invention is applied, and an ontology generating device according to a preferred embodiment of the present invention;

3 is a view illustrating a matching process between a language-based association word graph and an ontology in the language-based matching unit of FIG. 2 according to an embodiment of the present invention;

4 is a first view illustrating a matching process between a contextual information-based association word graph and an ontology in the contextual information-based matching unit of FIG. 2 according to an embodiment of the present invention;

FIG. 5 is a second view illustrating a matching process between a contextual information-based association word graph and an ontology in the contextual information-based matching unit of FIG. 2 according to an embodiment of the present invention; FIG.

6 is a view for explaining a semantic weighting process in the semantic weighting unit of FIG. 2 according to an embodiment of the present invention;

7 is a flowchart illustrating an ontology generation method according to an embodiment of the present invention.

* Explanation of symbols for the main parts of the drawings

10: ontology 20: association graph

30: ontology generation device 31: association word graph clustering unit

32: language-based matching unit 33: context information based matching unit

34: semantic weighting unit 35: inspection unit

36: forced matching unit 40: ontology storage

50: user terminal or semantic application

Claims (15)

In the ontology generating device, A language-based matching unit for matching language-based association information with an ontology-like system; A contextual information-based matching unit for matching the relationship matched by the language-based matching unit based on contextual information; And Weighting unit for assigning a semantic weight between the association information and the ontology-like system based on the relationship matched in the contextual information-based matching unit Ontology generation device comprising a. The method of claim 1, A clustering unit configured to receive pre-established related information and an ontology-like system and cluster the related information and deliver the related information to the language-based matching unit; A checker for receiving a matching relationship from the weighting unit, and checking whether a term consistently matches an ontology-like system in a clustering unit of the clustering unit; And Force matching unit for modifying the ontology-like system from the inspection unit or forcibly matching missing terms to the ontology-like system according to correction or supplementary information from an external expert. Ontology generation device further comprising. The method according to claim 1 or 2, The association information is an association word graph, The ontology-like system is an ontology, characterized in that the ontology. The method of claim 3, wherein The language-based matching unit matches target terms of the association graph with terms of the ontology by reflecting characteristics of language resources, The contextual information-based matching unit decomposes a graph of association words by generating a 1: 1 relationship by matching a relationship of 1: N (N is a natural number) generated by the language-based matching unit based on contextual information. The weight assigning unit assigns a semantic weight to a link relationship between an associated word graph and an ontology based on the relation decomposed by the contextual information-based matching unit. The method of claim 4, wherein The language-based matching unit, Match the terms of the ontology with the target terms in the association graph on a string-based basis, An ontology generating device for matching a term of an ontology and a target term of an association graph in a 1: N based on linguistic resources. The method of claim 4, wherein The situation information based matching unit, Match the relationship of 1: N (N is a natural number) generated by the language-based matching unit to the semantic-based by using the context information of the graph with the context information of the ontology, and the constraint-based Ontology generating device for matching and checking the connection relationship based on). The method of claim 4, wherein The weighting unit, An ontology generating device for assigning a semantic weight to a link relationship in the order of upper class <Link <attribute <subclass. The method according to claim 1 or 2, The association information is an association word graph, The ontology-like system is any one of a dictionary, a thesaurus, a taxanomi, or a folksonomy, ontology generation apparatus. In the ontology generation method, A language-based matching step of matching between terms of the association graph and terms of the ontology by reflecting characteristics of the language resource; A contextual information-based matching step of matching the matched relationship in the language-based matching step based on contextual information; And A weighting step of assigning a semantic weight to a link relationship between an associated word graph and an ontology based on the matched relationship in the contextual information-based matching step Ontology generation method comprising a. The method of claim 9, A clustering step of receiving a pre-built association word and ontology and clustering the association word graph to the language-based matching step; Receiving a matching relationship from the weighting step and checking whether a term is consistently matched to an ontology in a clustering unit of the clustering step; And Forced matching step of modifying the ontology from the inspection step or forcibly matching the missing term to the ontology according to the correction or supplementary information from an external expert Ontology generation method further comprising. 11. The method according to claim 9 or 10, The language-based matching step, Matching between terms of the ontology and target terms of the association graph in a string-based manner; And Matching between terms of the ontology and target terms in the association graph in a 1: N based on linguistic resources Ontology generation method comprising a. 11. The method according to claim 9 or 10, The situation information based matching step, Generating a relationship of 1: 1 by matching the relationship of 1: N (N is a natural number) generated in the language-based matching step based on context information, and decomposing a related word graph. 13. The method of claim 12, The situation information based matching step, Matching the 1: N relationship (N is a natural number) generated in the language-based matching step to semantic-based by using context information of a graph associated with context information of an ontology; And Identifying and matching the semantic-based matched relations on a constraint-based basis; Ontology generation method comprising a. delete delete

Images