Research And Implementation Of Words Similarity Model Based On Semantic

Words similarity computing based on semanteme, a question of much essential andimportant in automatic information processing, is widely applied in areas of informationretrieval，machine translation，QA systems，text mining and etc．Now there are variousmethods of words similarity computing, but results are not accurate because manyfactors, such as relations between words, are not considered in these methods.Based on 《synonyms dictionary》, which is developed by information retrieval labof HIT, structure information and principle of collecting words of 《synonymsdictionary》are fully analyzed, the influence of relations between words and worddistribution areas on words similarity computing has been studied in this paper. Newalgorithm on words similarity computing, through quantitative analysis on all thesefactors by means of experiment, is proposed and carried out in this paper.Three different methods of experiment are used to verify the rationality ofalgorithm, and also a comparative analysis from words similarity computing based on《How Net》(a thesis raised by Mr. Liuqun of Chinese Academy of Sciencescomputational place)is made in this paper. The algorithm is tested from the followingthree aspects：1. analysis on word alternative、2. experiment on statistical distribution ofword similarity、3. statistic analysis on synonyms, and comparative analysis of twomethods on rationality and accuracy have been made in this paper. As is shown in theresearch, the semantic similarity computing is efficient.This research, valuable and withgreat application prospect, can contribute to many domains in automatic informationprocessing.