Research On Concept Similarity Calculation Method Based On Ontology Mapping

Because of the extensive application of ontology and distribution characteristic of the World Wide Web (www), different users in the same field construct different onotologies which identify the common stock of knowledge. The ontology builded in different ways is called heterogeneous ontology. Ontology heterogeneity makes different ontologies identifing the same resource can not be reused and shared, and it is one of the main obstacles to mutual understanding, information exchange and interoperability between systems. Currently, ontology mapping is one of the principal ways for ontology heterogeneity. Ontology mapping is a process founding the links between two concepts. It is a normative statement of concepts consistency and relations consistency in ontologies. Ontology mapping framework includes five modules, in which concept similarity calculation is the key step, and the concept similarity algorithm directly affects the accuracy of the whole mapping result.This paper designs a comprehensive algorithm for concept similairy computation, considering from the sememe description of the concepts (DEF), instances and properties of concepts. The aim of the algorithm is to overcome drawbacks of traditional algorithm: time complexity and space complexity high, not comprehensive, error big and result not to be quantified. According to Hownet, concept can be explained by sememe, and the hierarchical architecture and taxonomy tree of sememes can reflect the relations between sememes. The paper calculates sememes similarity in two ways:main signalment and subordination signalment of sememes. The sememes which have no upper and lower relationships are calculated depending on their order number in the hierarchical architecture. The sememes which have upper and lower relationships are calculated according to the distance between them in the taxonomy tree.Similarity calculation based on sememes eliminates the ambiguity of concepts in the specific areas. But concepts belonging to different areas may be also having different meanings. So in the process of concept calcution needs some factors that can reflect concept location. In the ontology, instance and attribute are defined as a particular concept, and this definition is based on the area of the concepts. Based on the above discussion, the paper computes concept similarity from instances and attributes.The paper using statistics idea random extraction a part of instances from a mass of instances in the ontology designs an instances similarity algorithm. The algorithm matches similarity from instance range,instance scope and unit and uses condense matrix to storage intermediate data. Based on the frequently-used attribute relationships: functional properties,transitive properties,symmetric properties,inverse properties and the domain and range of attributes this paper designs an attribute algorithm. Nonetheless, for the concepts with the four relations in the same ontology, the paper uses the concepts which have a direct relationship mapping at the top of the concept hierarchical tree.At the end this paper, two heterogeneous medical ontologies are consturcted manually. By the heterogeneous medical ontologies, the algorithm is analyzed and verified, and compared the experimental data with the ontology mapping model MOMF result. In the appendix, the paper gives part of codes and data files by screenshots.