Research On Large-Scale Ontology Block Matching And Evaluation Based On Mixed Clustering

Ontology mapping aims to construct semantic bridge between different ontologies, through which the goal of knowledge sharing and information construction between different ontologies can be achieved. For the large-scale ontologies, it is more difficult to map them because of their characteristics, such as having a large number of concepts and complicated relations. This thesis focuses on this issue.Firstly, the research background of the semantic web, the purpose of researching and developing SNAX is briefly introduced, after which the state of the art of the ontology mapping technology is introduced , as well as the work about partitioning of large-scale ontologies.Secondly, according to the structure and features of large-scale ontology, a method of automatic clustering and partitioning for the class-hierarchy ontology based on the multiple clustering technology is proposed by applying Vector Space Model(VSM). It first constructs the vector space by using the words stemmed from the information of concepts, then represents the concepts in the vector space by using Semantic Diffuse algorithm, and then implements the multiple clustering algorithm on the vector space, after which the purpose of automatic clustering for the concepts is achieved. At last, the final mapping pairs are extracted according to the similarity value among the concepts in each certain mapped block pairs.Thirdly, as there are no evaluation systems for the automated partitioning technology in the domain of large-scale ontology mapping, this thesis designs a series of evaluation metrics, i.e., external evaluation and inner evaluation to measure the partitioning quality of the proposed method, considering the characteristics of the automatic clustering partitioning process.Finally, the partitioning and mapping system for the large-scale ontologies is designed and implemented. Moreover, the testing data sets of russia12 and tourismAB in the real life are used to test the mapping effect of the designed system, after which the strategies of POP and Rand are adopted to evaluate the partition effect of the large-scale ontologies.At last, the effect of the overall mapping is measured by computing the precision and recall value which is widely used in the information retrieval field. The results show that our method is comparatively good.