A Multiple Classifiers Integration Method Based on Full Information Matrix.pdfVIP

1000-9825/2003/14(06)1103 ©2003 Journal of Software 软 件 学 报 Vol.14, No.6 ∗ 基于全信息矩阵的多分类器集成方法 + 唐春生 , 金以慧 (清华大学 自动化系,北京 100084) A Multiple Classifiers Integration Method Based on Full Information Matrix + TANG Chun-Sheng , JIN Yi-Hui (Department of Automation, Tsinghua University, Beijing 100084, China) + Corresponding author: Phn: 86-10ext 3140, E-mail: tangchunsheng@; tangcs@ Received 2002-05-24; Accepted 2002-08-14 Tang CS, Jin YH. A multiple classifiers integration method based on full information matrix. Journal of Software, 2003,14(6):1103~1109. /1000-9825/14/1103.htm Abstract: Automatic text categorization is an effective method to increase the efficiency and quality of information utilizing. The combination of a set of different classifiers can often achieve higher classification accuracy. The concept of full information matrix is first given, and then an integration method of multiple classifiers based on adaptive weight adju sting is presented in this paper. The classifiers and their weights are determined automatically and adaptively with this method. The effective integration of each classifier ’s result can be realized by analyzing the statistical information of the classifier on the training set. The classification performance is promoted by the improvement of the precision and the recall. The effectiveness of the method is shown by the text classification experiments on the Reuters-21578 text sets. Key words: combination of multiple classifiers; full information matrix; text classification 摘 要: 自动文本分类是提高信息利用效率和质量的有效方法, 而多分类器的有效组合能够得到更高的分类 准确率.给出了样本集在多分类器下的全信息矩阵概念,并提出一种权重自适应调整的多分类器集成方法.该方 法能够自适应地选择分类器组合及确定