Chinese Learning of Semantical Selectional Preferences Based on LSC Model and Expectation Maximization Algorithm

摘要

Aiming at the situation of current Chinese language resources shortage ,this paper proposes semantically selectional preferences of unsupervised learning method, and presents a strategy of obtaining verb-noun semantic collocation in Chinese. An approach of Chinese semantic preference learning, which is based on Latent Semantic Clustering model and Expectation Maximization Algorithm. First, the parameters are initialized randomly. Second, a certain number of training iterations is performed until convergence. Each iteration consists of expectation step and maximization step. Finally, the semantic association between verbs and nouns are calculated as a measure of its matching probability. This method can be used on Chinese without syntax-annotated corpora. Lots of experiment results show that LSC provides proper patterns of verb-noun collocation semantically. The algorithm converges quickly.