Research On Concept Similarity Calculation Method In Ontology Mapping

As an important way to solve the problem of semantic heterogeneity of networkresources, the ontology construction method has no uniform norms and standards, andpeople develop new ontologies constantly, which leads to the problem ofheterogeneous ontology. Ontology mapping is one of the means to solve this problem.The key process is to calculate the concept similarity in order to measure the degreeof similarity between two concepts, and then establish semantic links. Many ontologymapping and concept similarity calculation methods have been proposed. Thesemethods have some problems, such as large amount of calculation, and poor results.On these issues, the main research work presented in this thesis is as follows:Firstly, ontology mapping technologies and concept similarity algorithms arestudied. There exist some issues in common similarity algorithms such as thesimilarity factors aren’t comprehensively considered and computation is complex. Sothis thesis introduces a semantic similarity calculation method based on semanticdistance, coincidence degree and node level difference. After analyzing impreciseproblem of distance calculation, a revised method is proposed. The revised methodassigns weights according to density of node, depth of node and type of edge in orderto improve the accuracy of the similarity. By using the WordNet semantic dictionary,a method which can locate two concepts of heterogeneous ontologies in the dictionaryis put forward, and the improved semantic similarity algorithm is applied to calculatethe concept similarity in WordNet.Secondly, for the issues about concept features not being fully taken into account,large computation resulting in inefficiency and low versatility in the existing mappingmethods, the thesis presents an ontology mapping model based on comprehensivesimilarity calculation. It uses the improved semantic similarity algorithm based onWordNet to derive the name similarity and extract the mapping candidates to reducecomputation, then calculate the similarity of mapping candidates based on attributes,structure and instances, and output the final mapping results after compositingvarious similarities. Finally, a simple testing system is designed to do comparative analyzingbetween improved method and unmodified method in precision, recall, F-measure,using the same test data set of OAEI(Ontology Alignment Evaluation Initiative). Theresults show that the improved method has higher accuracy and can distinguish thenuances among concepts better; the mapping model proposed in the thesis can reducecomputation significantly and has better performance.