Abstract: An improved twolevel text classification method is proposed, based on feature extraction. First, the characteristics of the text were extracted, and the weights were calculated. Then, the text was represented as a vector composed of characteristics and weight value. The vector angle cosine was used to calculate the similarity among the text so as to position the vast amount of information more accurately and rapidly. The experimental results show that the proposed classification method is superior to the existing center classification method in accuracy of classification, improving the adaptability and classification ability of the system.

Key words: text classification; feature extraction; vector space model; KNN algorithm