Hits:

Indexed by:期刊论文

Date of Publication:2008-04-01

Journal:Journal of Computational Information Systems.

Included Journals:EI

Volume:4

Issue:2

Page Number:495-501

Abstract:The high-dimension and sparseness of text data declines the performance of text classifiers. Consequently, it is vital for text classification to reduce the dimensionality of feature space effectively. In this paper, we propose an effective feature reduction method to address the problem. Firstly, we get rid of insignificant features using filtering method based on categorical document frequency distribution difference (CDFDD), and then conduct feature selection method based on variable precision rough set (VPRS) to obtain features significant to document category. To verify the effectiveness of our proposal, we adopt radial basis function (RBF) neural network classifier in experiment and test the classification performance by holding different number of features. Experimental results show that the proposed method outperforms Information Gain and X2 Statistic methods. What's more, it can achieve high precision and recall in condition of removing as much as 98% features.