Abstract: This paper compared several feature selection methods in text categorization, and proposed a new feature se-lection method(TF)DF Ci) based on weighted frequency of distinction between the text. It improves TF)DF function from weighted freduency and the feature items can increase the ability of text categorization in documents. In the experiment, we tested the effect of this feature selection method and other feature selection methods by using KNN classifiers. The experiments show the new method has good performance and stability under different numbers of training sets.

Key words: Feature selection,TFIDF,KNN classifiers