Abstract: Feature selection is the important factor which affects the question classification of question answering system. By fully using the characteristics of Chinese FrameNet in terms of semantic expression, this paper presents a new question classification-oriented approach in feature selection called strong class information words (SCIW). Firstly, it selects five kinds of Chinese FrameNet features as candidate features, and then uses SCIW to select features. According to each category’s classification precision of features, it sorts the classification ability of each single feature. Through the experiment of combinations of features, it selects the combination of features, which has better classification results. So the feature reduction can be reached.

Key words: Chinese FrameNet, question classification, feature selection