Study Of Ontology Mapping Based On Bp Neural Networks

In the fields of computer science, ontology is a formal specification of a shared conceptualization. The main purpose of applying ontology is knowledge sharing and reusing. Because there is not a common criterion for building ontology, so great deals of different ontologies are constructed in the same domain. These ontologies are heterogeneous. Ontology mapping can solve the problem of ontological heterogeneity.The key of ontology mapping is the computation of similarities. In the Semantic Web background, research areas such as ontology mapping,information integration and semantic search all need to recognize and estimate the semantic between different ontologies in order to share knowledge. However, currently, most exist technologies have some shortage and are not sufficient for applications technologies because these method sare depending on syntactic but not implied semantic.Under the background stated above, the analysis of exist approaches of concept similarity presented in this paper, proposing an integrated approach based on instances, definition, name and hierarchy information of concepts is presented. In this approach the instance similarity can be defined as joint probability distribution of the concept instances. And definition similarity can be got by statistics and graph theory. Meanwhile hierarchy similarity can be obtained by calculating the similarity of their corresponding semantic neighbourhood. Name similarity can be got by edit distance. And using BP neural network can get the weight of the four parts. Then the four parts can be integrated into the semantic similarity of two concepts.Do a simulation experiment based on a comprehensive calculation of the semantic similarity, training samples by repeated BP neural network, establishing the network output and expectations of the error signal output, adjusting the weight of network continuously, as much as possible to adapt to the input signal samples of the process. For the trained BP neural networks, testing with testing samples.It proves that the approach is effective and adaptive. It can be used to calculate the semantic similarity between two concepts. According to a single component to speculate concept of similarity between the semantic relations is not sufficient and biased. And the integrated similarity is more precise than any part of the three ones. BP neural network has the ability of self-learning and adjust the weight and threshold through learning and training the sample data. At the same time it still works in real scenarioes by using BP neural network to modulating the network weights without changing the similarity calculating model.