Research Of Hownet Based Word Semantic Computation And Application

Word similarity, relevance computation is always the basis of many fields such as Machine Translation, Sense Discrimilation and Information Retrieval etc. The paper introduces the word similarity, relevance computation situation in the world and analyzes several representative computation methods on word similarity, word relevance. It also gives the theoretic foundation and computation features.We propose the shortage of word similarity computation method based on HowNet recently and improves its computation formula. Then it can make more precise computation on word pairs with converse and Antonyous meaning. We also offer the computation methods on unlisted words and compound words which are never resolved by HowNet.Here we parse the hierarchy relation of concepts in HowNet and on the basis of it we design a concept-sememe tree structure, which describes the relation between sememes in concept, and it is easier to make researchers understand directly how the semantic description and computation HowNet does. HowNet doesn't provide the interface of word relevance computation so we make discussion and do research on HowNet resource, proposing word relevance computation formula on basis of HowNet, the experimental result shows the relevance computation method mentioned here can obtain satisfying results.We discuss the sentence similarity computation based on HowNet and we apply it to security FAQ system. Due to semantic computation, the system offers users expanding query function which can denote users wherether they care about other related questions even if the target question is not compiatble with question set by keywords.