特征选择及其在文本自动分类中的应用  

Feature Selection and Its Application in Automatic Text Categorization

作　　者：郝晓燕[1] 常晓明[1] 

机构地区：[1]太原理工大学计算机与软件学院,太原030024

出　　处：《电脑开发与应用》2006年第12期17-18,21,共3页Computer Development & Applications

基　　金：国家863高技术研究发展计划项目(2004AA115460)资助

摘　　要：比较研究了中文文本分类中四种不同的特征选择统计方法对k近邻分类器分类性能的影响。这四种特征选择采用的统计方法是:卡方、信息增益、互信息、交叉熵。相应地得到四种不同的特征集合。据不同的特征集合,分别进行了基于特征词布尔值和基于特征词的词频的中文文本分类实验,提出了文本分类系统流程,并给出了评估方法和实验结果。In this paper, the affects from four kinds of feature selection statistics methodologies to the classifying performances of k nearest neighbor categorizer are researched. The four kinds of feature selection statistics include Chi-square, information gain, and mutual information and cross entropy, and the four corresponding feature sets are obtained. According to the different feature sets, this paper carries out some experiments in text categorization respectively under Boolean value condition and under adding the frequency of words, and puts forward the system flow of text categorization, evaluation methods and experiment results.

关 键 词：文本分类 K近邻 统计方法 卡方 信息增益 互信息 交叉熵 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]