文本分类中一种基于核的最大散度差特征抽取方法

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。

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

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 result show that this method has a better precision.