Applying cascaded feature selection to SVM text categorization

This paper investigates the effect of a cascaded feature selection (CFS) in SVM text categorization. Unlike existing feature selections, our method (CFS) has two advantages. One can make use of the characteristic of each feature (word). Another is that unnecessary test documents for a category, which should be categorized into a negative set, can be removed in the first step. Compared with the method which does not apply CFS, our method achieved significant good performance especially about the categories which contain a small number of training documents.