对称和非对称词语聚类模型的比较研究-武汉大学学报·信息科学版.pdf

35 10 2009 5 Vol.35 No.10 Computer Engineering May 2009 ·· 2009 A TP391 ( 300072) 2 Comparative Research on Symmetric and Asymmetric Word Clustering Models SUN Yue-heng, CAO Gui-hong, HOU Yue-xian (School of Computer Science and Technology, Tianjin University, Tianjin 300072) AbstractWord clustering is one of important natural language processing issues in speech recognition and intelligent information retrieval, etc. This paper presents a symmetric clustering model based on mutual information. For the model not taking the order of words into account, it proposes a new asymmetric clustering model including two sub models, conditional clustering model and predictive clustering model. Experimental results on large scale data set show that compared with the symmetric clustering model, the asymmetric clustering model is a more effective one for clustering words. Key wordsword clustering; symmetric clustering model; asymmetric clustering model 1 k w c p (w | w w w ) p(w | c ) p(c | c c c )= 1 k n i i r k 1 2 k −1 k k k 1 2 k−1 n-gram cn − 1+ V − C V p (c