I. Introduction
With the wide application of Internet, network information resource is growing. It is difficult for people to quickly and effectively obtain the necessary information from large quantities of information. In order to quickly help users to find information and effectively use the information, it needs to categorize information organization management. Among them, the text information represents a significant component in the network information resources, so the research on text categorization technology is particularly important. Support Vector Machine is a new tool for solving the problem of machine learning based on optimization method, which originated in the early 1990s and put forward by Vapnik [1], acquired a breakthrough progress in theoretical research and algorithm recently. Many researchers put forward some deformation algorithm of the support vector machine. For example, v-SVM, Single category SVM, reduced SVM, weighted SVM and LS-SVM etc. These deformation algorithms are mainly through increasing function and variable or coefficient for formula deformation, producing the various kinds of one advantage algorithm or certain application algorithm [2]. Although people put forward many deformation algorithms in recent years, but it possess some features of great classification and many numbers of sample, much noise, disproportion of all kinds of numbers of sample, its slower speed and lower efficiency when we apply these algorithms to classification of those texts. We introduce the theory of rough set to overcome bottleneck problem of support vector machine.