Method for feature word weight calculating

Automatic text categorization has been one of the hotspots in the information processing field. To aim at the important impact of feature weight calculating on text classification accuracy, first, the relationship between text representation model and feature weight calculating is studied, and the existed methods of feature weight calculating are analyzed, then the common idea of feature weighting for vector space model (VSM) summarized. Second, the feature weighting thoughts for class space model (CSM) are gived, and on the basis of some existed methods, a number of feature weight calculating methods for CSM are proposed. Experimental results show that the proposed methods are effective in improving classification performance. After analyzing the experimental results, this paper points out that the appearance and the disappearance of feature words together to consider will lower the classification accuracy, so the favorable factors of a class should be highlighted, and the negative side should be ignored or minused.