一种用于中文主题分类的CSVM算法

摘要

This paper proposes a new Cascade Support Vector Machine(CSVM) classification algorithm, using AdaBoost algorithm framework and Support Vector Machine(SVM) combination to deal with the problem of multiple classifier. For the problem of consuming time in the multi-classification problems with SVM, introduces the Minimum Enclosing Ball(MEB) algorithm to extract the original sample data to shorten the training time for SVM. CSVM is applied in the Chinese thematic classification. Experimental results show that CSVM algorithm has similar accuracy with AdaBoost algorithm, but the computation time is only 35% of the AdaBoost-SVM algorithm.%提出一种新的级联支持向量机分类算法CSVM,结合AdaBoost算法框架与支持向量机(SVM)进行多分类处理.针对多分类问题中支持向量机处理样本数量多和计算时间过长的问题,引入最小闭合球算法对原始样本数据进行提取,以缩短SVM的训练时间.实验结果表明,CSVM算法具有与AdaBoost-SVM算法相似的精确度,而计算时间仅为AdaBoost-SVM算法的35％.