Word sense discrimination is to group occurrences of a word into clusters based on unsupervised classification method, where each cluster consists of occurrences having same meaning. Feature extraction method has been used to reduce the dimension of context vector in English word sense discrimination task. But if original dimension has a real meaning to users and relevant features exist in original dimensions, feature selection is a better choice for finding relevant features. In this paper we apply two unsupervised feature selection schemes to Chinese character sense discrimination, which are entropy based feature filter and Minimum Description Length based feature wrapper. Using precision evaluation and known ground-truth classification result, our preliminary experiment results demonstrate that feature selection method performs better than feature extraction method on Chinese character sense discrimination task.
展开▼
