中文文本分类器的评价  

Evaluation of classifiers for Chinese text categorization

作　　者：张杰[1] 战学刚[2] 冯金平[1] 陈文亮[3] 

机构地区：[1]中国人民解放军炮兵学院二系,安徽合肥230031 [2]鞍山科技大学网络中心,辽宁鞍山114044 [3]东北大学计算机软件所,辽宁沈阳110004

出　　处：《鞍山科技大学学报》2005年第3期231-234,238,共5页Journal of Anshan University of Science and Technology

摘　　要：对目前比较流行的4种中文文本分类器(Rocchio、KNN、NaiveBayes、最大熵)进行评价,其中,NaiveBayes和最大熵是基于概率统计的方法,而Rocchio和KNN是基于向量的相似度计算的方法。选用χ2作为文本特征选取方法,对一个中文文本分类语料库进行分类评测。实验结果表明,最大熵和NaiveBayes的分类性能十分接近,处于较好水平,而KNN和Rocchio分类性能稍差一些。Four different classifiers for Chinese text categorization was evaluated which were:Rocchio,KNN,naive Bayes,maximum entropy.Naive Bayes and maximum entropy are the classifiers based on probabilities.And Rocchio and KNN are based on the similarities between the vectors.Using χ~2 as feature selection method and conduct the experiments on a Chinese text corpus was carried out.The experimental results show that naive Bayes and maximum entropy are top performers in this evaluation.In contrast,Rocchio and KNN provide a lower performance.

关 键 词：文本分类 分类器 评测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]