Feature selection method combined optimized document frequency with improved RBF network

Abstract

Feature selection is the core research topic in text categorization. Firstly, it combined word frequency with document frequency and presented an optimized document frequency (ODF) method. Then it proposed an adaptive quantum-behaved particle swarm optimization (AQPSO) algorithm in order to train the central position and width of the basis function adopted in the RBF neural network. Next the weight of the RBF network was computed by means of least-square method (LSM). Finally, a combined feature selection method was provided. The combined feature selection method firstly uses the optimal document frequency method to filter out some terms to reduce the sparsity of feature spaces, and then employs the improved RBF neural network to select more outstanding feature subsets. The experimental results show that the combined method is effective. © 2009 Springer.