Abstract: Automatic text classification is defined as the task to assign pre-defined category labels to documents．Based on the limitations of Vector Space Model, the Vector Space Model is incapable of expressing the structure of documents effectively．To solve this problem，this paper constructs the sireilar matrix by train text, and achieves the subject information of each category through similar matrix, and then to construct the classifier by the subject information．Finally the classifier is combined with the classic classifier to determine the category of text．The experiment system
shows the effectiveness of the method．

Key words: text classification, semantic features, VSM, graphical model, algorithm