On-line text categorization algorithm based on information fusion: Semantic SVM

The aim of this paper is to make SVMs (Support Vector Machines) more applicable to on-line text categorization applications. As SVMs are of good generation ability even with small training sets and text feature vectors are clustery in the feature space, an algorithm for text categorization, namely, semantic Support Vector Machine (Semantic SVM), is proposed by substituting the original training text set with the semantic center set. This semantic center set is used as the training text and support vector candidates. The steps to generate the semantic center set and the framework of the on-line learning algorithm of semantic SVM are then presented, as well as the implementation of the on-line learning algorithm based on Sequential Minimal Optimization. Experimental results show that, compared with the standard SVMs, the proposed semantic SVM and its algorithm can improve the on-line learning speed and the classifying speed by orders with a high classifying veracity.