一种基于散度差组合型文本特征降维方法  

A Combined Method of Feature Reduction Based on Principal Scatter Difference in Text Classification

作　　者：刘海峰[1] 姚泽清[1] 王元元[2] 张述祖[1] 

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

出　　处：《微电子学与计算机》2008年第8期69-71,75,共4页Microelectronics & Computer

基　　金：国家自然科学基金项目(70571087)

摘　　要：讨论了文本分类中特征降维的主要方法及其特点,分析了基于散度差准则的特征降维的原理和方法,在避开求逆矩阵问题的同时,通过对文本特征进行选择对文本特征集进行了第一次压缩,借助于加权散度差原理对特征集进行了二次抽取,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降低.试验结果表明,这种方法在文本分类上的效率较好.The problem of feature dimension is one of the main problems in text classification. In the paper we bring forward a method of extracting the text features based on scatter difference and CHI statistics. Firstly, we analyse the primary feature reduction means and their characteristic in the text classification. Secondly, we analyse the principle and method of scatter difference criterion. Avoid calculating the inverse matrix, we realize the first feature reduction by using feature selection. Then we use a weighted-scatter difference to extract the spare features. At the precondition of lower information loss we reduce the feature dimension. Lastly, our test about text categorization shows 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 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]