Abstract: Text feature reduction is the key technology in text categorization. In addition, K-means is an partitioning method which usually be used. With regards to this arithmetic excessively incentive to the initial centers and the isolated points, the improved K-means arithmetic was put forward which is used in text feature selection. Text feature clustering was improved by optimizing primitive class center's options and the elimination of isolated point Following text classification test shows that the K-means arithmetic put forward in this paper has a good feature selection ability and high efficiency in text categorization.

Key words: Feature selection, Clustering, K-means, Text categorization