An Ontology Matching Approach Based On Selection And Fusion Of Multiple Strategies

With the rapid development of Internet technology, the knowledge distributed on Web is increasing with explosive growth. Toward sharing and integrating the knowledge on Internet, ontology is introduced as a generic framework to store as well as describe knowledge. In a specific domain, ontology formally defines acknowledged terminologies to express objects, therefore the separated knowledge on semantic web is able to be integrated into a homogenous unit. Nevertheless, due to the subjectivity of researchers and the decentralized nature of the Internet, when different organizations utilize ontology to model the knowledge in the same specific domain, ontology heterogeneity would emerge inevitably. For the purpose of solving such a problem, the ontology matching methods are extensively studied as a key technology.The thesis will focus on the research of ontology matching. To be specific, three aspects are involved:(1) The fundamental matchers applied in ontology matching are studied. Diverse matchers are established based on different ontology features, such as the name-based matcher, annotation-based matcher and structure-based matcher. Moreover, the word2vec model is studied and introduced into the matching system. On basis of the word embedding model, a name-based matcher and an annotation-based one are designed and presented.(2) The machine learning model is studied. The model is employed to select and integrate various individual matchers. In order to use matchers synthetically, the similarities that each matcher computes are treated as sample features to build training set. After learning the training set, a trained machine learning model is constructed. The established machine learning model is capable of selecting and integrating similarity metrics so as to establish an effective ontology matching system.(3) An ontology matching system is designed and implemented and a series of experiments are designed to test the matching system. The system is divided into different modules which perform distinct functions. Through the interactions of different modules, the alignment of heterogeneous ontologies is therefore achieved. Moreover, OAEI standard data sets are employed in the experiments to test the matching effect. The results of experiments indicate that the matching algorithm proposed in this thesis is capable of discovering matching entities between heterogeneous ontologies in an effective way. Additionally, based on the experiment results, the system is proved to have a relatively high accuracy of ontology matching.