基于MRMR的文本分类特征选择方法  

Feature Selection Method Based on MRMR for Text Classification

作　　者：李军怀[1] 付静飞 蒋文杰[1] 费蓉[1] 王怀军[1] LI Jun-huai FU Jing-fei JIANG Wen-jie FEI Rong WANG Huai-jun(Xi＇an University of Technology,Xian 710048,China)

机构地区：[1]西安理工大学,西安710048

出　　处：《计算机科学》2016年第10期225-228,共4页Computer Science

基　　金：国家自然科学基金(61172018);陕西教育厅科技计划(15JS077);西安市科技计划(CXY1439(8))资助

摘　　要：特征选择是文本分类技术中重要的处理步骤,特征词选择的优劣直接关系到后续文本分类结果的准确率。使用传统特征选择方法如互信息(MI)、信息增益(IG)、χ2统计量(CHI)等提取的特征词仍存在冗余。针对这一问题,通过结合词频-逆文档率(TF_IDF)和最大相关最小冗余标准(MRMR),提出了一种基于MRMR的特征词二次选取方法 TFIDF_MRMR。实验结果表明,该方法可以较好地减少特征词之间的冗余,提高文本分类的准确率。Feature selection is the most important preprocessing step in text classification The quality of the feature words has a significant impact on the accuracy of the classification results. Using the traditional feature selection method, such as MI, IG,CHI, will still cause the redundanly among the feature words. For this problem, based on the combina- tion of the term frequency-inverse document frequency （TF-IDF） and the maximal relevance minimal redundancy （MRMR） ,this paper put forward a MRMR based feature words secondary selection method TFIDF_MRMR. The ex- perimental results indicate that this method is able to reduce the redundancy of feature words and improve the accuracy of classification.

关 键 词：特征选择 最大相关最小冗余 词频-逆文档率 文本分类 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]