Research On Key Approaches Of Heterogenous Ontology Mapping

With the rapid development of the information technology era, the World Wide Web haschanged the way people communicate with each other, and the way of informationdissemination, access and commercial operation. Information resource is explosivelyincreasing, and it’s extremely difficult to accurately, quickly retrieve information from massdata. Because most of the information on the World Wide Web is expressed in a user readableformat and lack of semantic information, software agent is unable to understand and deal withthe information. In order to solve this problem, the inventor of the World Wide Web,T．Berners-Lee proposed the Semantic Web vision. The Semantic Web is the expansion andextension of the current Web, it aims to realize the data exchange, knowledge sharing andreuse between different information systems, namely semantic interoperability. But in thedistributed environment of the Semantic Web, it’s so difficult to realize semanticinteroperability. Ontology has become a key factor of the interoperability of heterogeneoussystems. It is the foundation of the Semantic Web and can describe the semantic informationof data.In order to achieve knowledge sharing and reuse, many different fields have definedcorresponding standards of ontology, but ontology construction does not have uniformstandard to restrain. In addition, ontology designers’ different understandings of concept,attribute, relationship and other elements of different fields lead to the emergence of theproblem of heterogeneous ontology. It impedes the semantic interoperability betweendifferent information systems. As the key to solve this problem, ontology mapping hasbecome an important research topic and attracted many research institutions and scholars.Ontology mapping can find semantic correspondences between elements of differentontologies.At present, although there are many research institutions and scholars have developedmany approaches and technologies related to ontology mapping, there is no one that cancompletely meet the needs of future development, and automatically complete all operationswithout the participation of domain experts. Therefore, the research of ontology mapping isfaced with many challenges.In this paper, the author does thorough and careful research onthe existing ontology mapping problems, and proposes three approaches of ontology mappingto solve practical problems. Firstly, approaches of ontology mapping based on multi-strategy are proposed. Themapping ontology usually contains a variety of information, such as lexical information,taxonomy structure, syntactic information, semantic information and constraint information,which can be regarded as ontology characteristics.Single ontology mapping approach can not use all information of all entities in twoontologies. Using the combined various strategies can make full use of ontology informationand produce better results than just using a strategy.Therefore, approaches based onmulti-strategy has been wildly adopted by most ontology mapping methods. In this paper, wepropose approaches of ontology mapping based on multi-strategy which mainly considerconceptual mapping and property mapping.The approach of conceptual mapping adopts conceptual similarity strategy based onconceptual name, property and taxonomy to calculate conceptual similarity. It not onlyconsiders semantic and lexical information of concepts, but also considers attributes andtaxonomy of concepts. The approach of property mapping adopts different strategies to dealwith data type properties and object type properties. Aiming at solving the problem of verylarge amount of computation in property similarity calculation between two ontologies, it’snecessary to restrict the number of property pairs. In order to produce propertymappings between two ontologies, we filter properties to get the most relevant properties asproperty candidate set for a property. Only for a property and property in its candidate set,similarity between them is calculated. After the mapping results based on multi-strategy arecombined, the best mapping results are chosen, and conceptual mapping and propertymapping are optimized. Due to the full use of information in the ontologies, our approachesavoid the drawback of single ontology mapping approach which can not utilize all theinformation in ontologies. We present the experimental results with high recall and precision.Our approaches based on multi-strategy have high mapping efficiency and accuracy.Secondly, we propose ontology mapping method based on conceptual candidate set. Atpresent, most methods calculate the similarities between all concept pairs in ontologies whencalculating conceptual similarities between two ontologies. That results in large amount ofcalculation. Some concepts are not similar at all, it is not necessary to calculatesimilarity between them. Therefore, it is necessary to limit the number of concept pairs inorder to reduce the amount of calculation. In this paper, we present an ontology mappingmethod based on conceptual candidate set. When calculating conceptual similarity, it not onlyconsiders semantic information of concepts, but also considers property and taxonomy ofconcepts.For a concept in an ontology, conceptual name similarity between it and all conceptsin another ontology is calculated. The threshold is set and the candidate set for this concept isproduced.Only for a concept and concept in its candidate set, similarities based on propertyand structure between them are calculated. Similarities are combined and the results of ontology mapping are produced. And experimental results prove that our method has higherrecall and precision.The amount of similarity computation is reduced by using candidate setfor concept, and high efficiency of this ontology mapping method is achieved.Finally, we propose an ontology partitioning and mapping method based onROCK clustering. With the expansion of ontology application fields, there are a variety oflarge scale ontologies defined by specialized organizations. Large scale ontologiesusually involve many fields. For the mapping task related to these ontologies, notall entities are related to mapping, and only concepts in relative fields may have mappings.But most of the existing ontology mapping methods do not consider the fields ofontology entities, and calculate similarities between all entity pairs. It will cause moremismatches and low mapping efficiency. With the wide application of large scale, multi-fieldontologies, it poses great challenges to the present mapping methods in the aspectsof mapping precision and mapping efficiency. ROCK clustering is applied in partitioning inthis paper. Two large scale ontologies are preprocessed and concept pairs are extracted. Thensimilarities based on semantics, substring and taxonomy are calculated, and we obtain linksbetween concepts. The results of links between concepts are more comprehensive becausemany kinds of ontology information are used. By calculating cohesiveness and coupling, twoontologies are partitioned into blocks respectively based on improved ROCK clusteringalgorithm. We adopt block matching strategies based on taxonomy and anchors, and obtainsimilarities between blocks and block matching results. And experimental results show thatour method has high mapping accuracy.The three ontology mapping approaches are compared below. The approaches ofontology mapping based on multi-strategy are aimed to solve conceptual mapping andproperty mapping. The approach of conceptual mapping adopts conceptual similarity strategybased on conceptual name, property and taxonomy to calculate conceptual similarity. Theapproach of property mapping adopts different strategies to deal with data type properties andobject type properties.Then conceptual mapping and property mapping results are optimized.The approaches fully make use of information in the ontologies, such as semantics, syntacticinformation, taxonomies and property information.Ontology mapping method based onconceptual candidate set limits the number of concept pairs in order to reduce the amount ofcalculation. It not only considers semantic information of concepts, but also considersproperty and taxonomy of concepts.The method for ontology partitioning and mapping basedon ROCK clustering are suitable for large scale ontologies. Two ontologies are partitionedinto blocks respectively based on improved ROCK clustering algorithm. Block matching strategies based on taxonomy and anchors are adopted, and block matching results areobtained.Although the experimental results on dataset show the effectiveness of these threemethods, there also exist some problems and shortcomings. Therefore, in future work, we willmake improvements in these methods in order to further enhance the performance.