Research On The Method Of Semantic Retrieval Based On Ontology

With the rapid development of science and technology in recent years, the Internet has become a main way for people to obtain information resources. For the mass information in the Internet, how to quickly and accurately obtain the required resource has become an urgent problem to be solved. Traditional retrieval system or search engine is mainly carried out by mechanical matching keywords, and it lacks of understanding and analysis of the semantic level for the query request as well as information resources, which causes the actual retrieval results do not match with the users query demand.As a good kind of concept modeling tool, Ontology not only can describe information in semantic and knowledge level, and reflect the relationship between the concepts, but also support logic operation. So ontology is introduced into the traditional information retrieval system. The semantic query is acquired by making use of the advantage of ontology.Accordingly, it improves the recall and precision. In this paper, after analyzing and discussing the current methods of semantic retrieval, the following three aspects are studied:(1) Based on the deep understanding of the distance-based, content-based and feature-based methods,a comprehensive weighted calculation method of semantic similarity is proposed. The semantic distance, semantic coincidence degree and attribute coincidence degree between concepts are comprehensively taken into consideration. In the proposed method, the influence of depth, refinement degree and type on directed edge weight are analyzed in the process of semantic distance similarity measure; the factors of depth difference between concepts are added in the calculation of semantic coincidence degree,which improves the problem that cant distinct the semantic coincidence degree between concept pairs when they have the same common upper nodes and the same sum of depth. In the aspect of attribute coincidence degree calculation, some concept nodes based on inheritance of the father’s concepts attributes are considered and new attributes are defined again. And direct and indirect calculation methods are given. The experimental results show that the improved algorithm can more accurately measure the similarity between the concepts.(2) For the problem that Ontology query expansion breaks away from retrieval documents, in this paper, a method of query expansion based on ontology and local Co-occurrence by combining the semantic expansion and the statistics expansion is proposed.In the ontology, some concepts which have high similarity with the original query concepts as semantic candidate concepts are selected. Similarly, the statistics candidate concepts are obtained by co-occurrence analysis in the local documents. Then the local documents and the ontology are respectively employed to filter semantic candidate concepts and statistics candidate concepts. After finishing the filter, the two parts of extension words are used as the final expansion concepts. The new query obtained from query expansion algorithm can reflectthe user’s request better.(3) Research on the problem that ontology-based query expansion seriously depends on the building domain ontology. A query expansion method based on similarity and relevance of concept is presented. The correlation between concepts is made up of concept similarity in the ontology and concept relevance in local documents. The measure of correlation between concepts does not only rely on ontology. And the expansion concepts according to the relevance between concepts and the whole initial query are selected. In addition, the calculation method of extension term weights is proposed. The weight is determined by concept weight in the local documents, the similarity in the ontology and the relevance in the local documents. In the design of retrieval system, the proposed algorithm can effectively improve the retrieval performance.