Research On Domain Ontology Concept Extraction And Relation Extraction

With the development of society, the demand for digital information becomes stronger. Information technology is facing challenges such as information representation, knowledge sharing, software reuse, etc. It arouses wide public concern that how we can do well in massive information's organization, management, maintenance and reuse on the network,and how to provide effective and rapid services for business users. As a shared conceptual model, Ontology has become increasingly attractive. It has been widely used in many areas such as knowledge engineering, artificial intelligence, semantic retrieve and so on. But the manual construction of ontology is a time-consuming task: it takes considerable time and resources, especially, the construction of domain ontology needs experts'participation. Therefore, the construction of ontology leads to the bottleneck of knowledge acquisition. To solve the problem, people began to try to automatically or semi-automatically construct ontology, that is ontology learning.Ontology learning obtains expectant ontology in semi-automatic or automatic way from existing data sources. Always machine learning, statistics, natural language technology can be used. Currently, research on this area focuses on concept extraction and relation extraction. Statistics-based approach is mainly adopted in traditional ontology learning, but ontology concepts and relations are more focused in the aggregation of semantic. Because the impact of semantics has been ignored, there are many substantial inaccuracies in learning result.To solve above problems, an ontology learning method based on filtering mechanism has been proposed in this paper. This method uses the context of concept to construct vector space model, and then get semantic similarity to represent the semantic relevance between words. At last, semantic similarity is used to filter concepts and relations that has been extract before.Additionally, research on classification relations has been presented. In this paper we use the method of term includes to get classification relations, and use computing formula to measure the classification relations.In order to verify validity of the model, this paper constructs an ontology learning system. This system compares traditional ontology learning method with ontology learning method based on semantic filtering. To enhance the objectivity of experimental evaluation, Hownet similarity calculation software is used to construct contrasting ontology. Experimental result shows that the improved model can effectively improve precision of concept extraction and relation extraction, and verify ontology learning method based on semantic filtering is effective.