A Knowledge Based Method for Chinese Word Sense Induction

Word sense induction is usually viewed as a cluster problem in natural language processing. The context of the target word is represented as a vector and the cluster algorithms such as k-means, EM are applied. Different from the traditional methods, we proposed a new way based on “one sense per collocation” assumption which is proposed by Yarwosky (1993). Each sentence which contains the polysemous words is first parsed by Stanford parser, in order to find the collocation word of the polysemous word. Then, according to the collocation words´ semantic category, the sentences are divided into different clusters. The experiments were run on the benchmark data set, and the results show the effect of the method.