A hybrid model for sense guessing of Chinese unknown words

This paper proposes a hybrid model to address the task of sense guessing for Chinese unknown words. Three types of similarity, i.e., positional, syntactic and semantic similarity, are analyzed; and three models are developed accordingly. Then the three models are combined to form a hybrid one (HPPS Model). To verify the effectiveness and consistency of HPPS, experiments were conducted on ten test sets which were collected from two popular Chinese thesauruses Cilin and HowNet. In addition, extra experiments were made on a test set of 2000 words which were collected from newspaper. The experiments show that HPPS Model consistently produces 4%~6% F-score improvement over the best results reported in previous researches. © 2009 by Likun Qiu, Kai Zhao, and Changjian Hu.