关键词: 类比检索,SVM,语义相似

Abstract: With the development of Internet,the problem of not acquiring information of unknown domains because not exactly import keywords becomes more common．As a new retrieve method of acquiring knowledge of unknown domains using the knowledge of known domains,analogy retrieval gradually becomes one of hot topics．Analogy retrieval first analyzes the potential relationships between pairs of words and then accurately returns target information using these relationships．For example,given an analogy query Q={A:B,C:?},here it is assumed that there are some potential relationships between A and B.The aim of analogy retrieval is to determine the target(s) D of ?,and the relationships between two pairs of words,A and B,C and D,are similar．Two key difficulties of analogy retrieval are:(1) mining relationships between two words and (2) extracting target words．Both of them are more challenging in Chinese．This paper proposed a SVM based Chinese Analogy Retrieval (namely SVMbCAR) with two main components,SVM based relation-words extracting and SVM based target words determining．Experiments on a real-life data set (600person entity pairs from Ren Li Fang) show that the accuracy of extracting relationships between two words is 82.3%,and the accuracy of extracting target words is 90.5%.

Key words: Analogy retrieval,SVM,Semantic similarity