基于X^2统计的改进文本特征选择方法  

Improved method for text feature selection based on CHI

作　　者：闫健卓[1] 李鹏英[1] 方丽英[1] 龙丽英[1] 刘欣悦[1] 

机构地区：[1]北京工业大学电控学院,北京100022

出　　处：《计算机工程与设计》2016年第5期1391-1394,共4页Computer Engineering and Design

基　　金：国家自然科学基金项目(61174109);北京市委组织部优秀人才培养计划基金项目(2010D005015000001)

摘　　要：为克服传统X^2统计模型未考虑特征词频数因素不足的缺陷,根据特征项的词频数及其在类间、类内不同分布情况,提出一种改进的X^2统计特征选择方法,使特征频数信息得到有效利用。实验对比改进前后的方法对文本分类的结果,实验结果表明,改进后方法的分类效果优于传统X^2统计方法,验证了其有效性。Traditional X^2 statistical model fails to consider the frequency of the feature terms,an improved Chi-square statistic（CHI）algorithm based on frequency and its distribution within class and between classes was proposed to make full use of the frequency of features.The experimental results of text categorization using the improved method were compared to that of other methods.Results of analysis indicate that the proposed algorithm is better than the traditional method and verifies the effectiveness of the proposed method.

关 键 词：文本分类 特征选择 X2统计 特征频数 特征分布 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]