Ontology Matching Based On Markov Logic Network

Ontology matching is the process of determining the correspondences between concepts, properties, and instances of different heterogeneous ontologies, to achieve the goal of knowledge base from different domains sharing and information exchanging.It has widely been known that logical semantics and reasoning are the basis of intelligent applications on the semantic web. However, Existing ontology matching systems focus on the similarity of semantics, the use of logical semantics as a reasoning method has long been neglected. With this paper, we will focus on the research about the reasoning ability of logical semantics, proposes a novel probabilistic-logical framework based on Markov logic network combining the similarity of semantics and the uncertain reasoning.Firstly, we will briefly introduce the background and significance of the research, and then analyze the existing ontology matching systems from the similarity and reasoning respects, respectively. By summarizing the advantages and limitations of these methods, meanwhile, pointing out the common defect of the multi-strategies matching systems, we will point out the direction and goals of our work.Secondly, with presenting the markov network and first-order logic, we propose the method of based markov logic network ontology matching. Compared to existing systems, our method can make best use of the property of first-order logic, combining the hard formula and soft formula. It first employs multi-strategies to get the similarity between concepts, properties, and instances of different ontologies. Then, then, take the synthesized similarity as a prior probability for dealing with the uncertainty bringing by the imprecise semantic information, namely, the soft formula of first-order logic; on the other hand, consider the known logical statements—hard formula, reduce the incoherence of knowledge base.Markov logic network combines first-order logic and undirected probabilistic graphical models. MAP estimation in Markov logic network is NP-complete, therefore, we exploit the Integer Linear Programming (ILP) express MAP (maximum a posteriori probability) problem, then, based on the ILP Base solver apply Cutting Plane Inference (CPI) to find mappings, which was shown to be an effective method for exact MAP estimation in undirected graphical models.Finally, based on the research above, we designed and implemented the matching system based on Markov Logic Network. We use the dataset offered by OAEI as basis for our experiments. Comparing with the other matching systems, we show empirically that the approach is efficient on exploiting deeper matching relationship,and it is more accurate than existing matchers on ontology matching.