基于散度差准则的文本特征降维研究  

Research of reducing text feature based on scatter difference criterion

作　　者：刘海峰[1] 王元元[1] 张学仁[2] 刘守生[2] 

机构地区：[1]解放军理工大学指挥自动化学院,解放军理工大学理学院南京210007 [2]解放军理工大学理学院,南京210007

出　　处：《计算机应用研究》2008年第7期1971-1973,共3页Application Research of Computers

基　　金：国家自然科学基金资助项目(70571087)

摘　　要：研究了一种基于散度差准则的文本特征抽取方法。首先讨论了文本分类中特征降维的主要方法及其特点,然后分析了一种基于散度差的准则用于特征降维的原理和方法,从理论上对该方法的相关步骤进行了数学论证。在中文文本分类实验中,对KNN分类器进行了基于密度的改进,消除了由于文本分布倾斜对分类器产生的影响。实验结果表明,这种方法在文本分类的准确性方面效果较为理想。The paper studied a method of extracting text feature based on scatter difference. Firstly, analyzed the primary feature reduction means and their characteristic in text classification. Secondly, analyzed the principle and its method that based on scatter difference criterion. And more, the paper demonstrated the validity about the main approaches of this method. Lastly, had a test about Chinese text categorization with this way. It improved the KNN classifier in the density factor to eliminate the data incline disadvantage. The result shows that this method has a better precision.

关 键 词：文本分类 特征选择 特征抽取 特征降维 散度差 KNN分类器 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]