Word Similarity Computing Based On Chinese Conceptual Graph

As a key point in Chinese information processing, the calculation of semantic similarity has been researched widely and deeply by a lot of researchers. It is the basis of many important fields such as information retrieval, information extraction, text classification, word sense disambiguation, machine translation. There are two kinds of methods: one is rule-based and another is corpus-based. However, all these methods mainly rely on two words'distance in semantic dictionary or their correlation obtained from corpus without considering Chinese connotations.This paper presents a new method for semantic similarity calculation, which focuses on the perspective of connotations, trying to highlight the essential attributes of words by which we can get the similarity value in words'conceptual level. This method transforms words into conceptual graph by using their definitions. The value of semantic similarity is obtained by calculating the similarity of two conceptual graphs. The mainly contribution of this paper is:Firstly, this paper presents a method to construct connotation conceptual graphs which consists of four steps: obtaining interpretation items, analyzing concepts, extracting knowledge and constructing the conceptual graph.Secondly, this paper presents a semantic similarity calculating method based on the sememe according to the definitions of words. This method is the basis of calculating similarity of conceptual graph, which will be used to calculate the similarity between nodes.Thirdly, with the similarity of nodes obtained, this paper presents a method to measure the similarity between conceptual graphs. We extract amounts of attribute names which can completely represent the words'connotation in certain field. Then a recursive algorithm is called to calculate the whole similarity of two conceptual graphs.At last, the similarity method is applied to the web classification to verify its validity. And the evaluations show that the method is effective and achieves good results.To sum up, this paper presents a new method for the semantic similarity calculation which provides effective technical support for new generation of search engines by analyzing the similarity in conceptual level. It is an important component of language engineering.