Abstract

Document categorization has become one of the most important research areas of pattern recognition and data mining due to the exponential growth of documents in the Internet and the emergent need to organize them. The document space is always of very high dimensionality and learning in such a high dimensional space is often impossible due to the curse of dimensionality. To cope with performance and accuracy problems with high dimensionality, a novel dimensionality reduction algorithm called IKDA is proposed in this paper. The proposed IKDA algorithm combines kernel-based learning techniques and direct iterative optimization procedure to deal with the nonlinearity of the document distribution. The proposed algorithm also effectively solves the so-called “small sample size” problem in document classification task. Extensive experimental results on two real world data sets demonstrate the effectiveness and efficiency of the proposed algorithm.