文本分类中一种基于核的最大散度差特征抽取方法  

Kernel-based maximum scatter difference method of feature extraction in text classification

作　　者：刘海峰[1,2] 姚泽清[1] 张述祖[1] 王元元[2] 

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

出　　处：《计算机应用研究》2009年第1期102-104,107,共4页Application Research of Computers

基　　金：国家自然科学基金资助项目(70571087)

摘　　要：研究了一种基于核的最大散度差准则的文本特征抽取方法。首先回顾了文本分类中特征降维的主要方法、Fisher准则及其相关研究进展以及存在的问题;然后分析了基于散度差准则的线性鉴别方法的优点与不足,借助于核函数较好地解决了线性可分性较差的样本分类问题,在最低限度减少信息损失的前提下实现了特征维数的大幅度减缩。实验结果表明,该方法在文本分类上的效果较好。This paper studied a method of extracting the text features based on the kernel and scatter difference. Firstly, reviewed the primary feature reduction means, Fisher discriminant and the developing of this aspect. Secondly, analyzed the virtue and the defect of the scatter difference criterion. In virtue of the kernel it solved the stylebooks classification problem that had less linear separability. At the precondition of lower information loss, reduces the feature dimension greatly. The lastly, a test about text categorization and the resuh show that this method has a better precision.

关 键 词：文本分类 特征抽取 特征降维 散度差准则 核变换 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]