融合全局和局部特征的文本特征提取方法研究

Study on Text Feature Extraction Method with Integration of Whole and Part Features

[目的 /意义]为了解决主流特征提取方法的提取效率有限的问题,在Fisher判别分析的基础上,借鉴流形学习思想,提出融合全局和局部特征的文本特征提取方法。[方法 /过程]首先定义基于流形的类间离散度(MBCS)以及基于流形的类内离散度(MWCS),然后在Fisher准则基础上通过最大化MBCS与MWCS之比实现特征提取,从而保证类间样本尽可能远离,而类内样本尽可能紧密。[结果 /结论]比较实验结果表明该方法有效。

[Purpose/significance] To solve the problem of limited extraction efficiency of current text feature extraction methods, the paper takes manifold learning for reference on the basis of Fisher Discriminant Analysis （FDA）, and puts forwards a text feature ex- traction method with integration of whole and part features. [Method/process] The paper defines Manifold-based Between-Class Scatter （MBCS） and Manifold-based Within-Class Scatter （MWCS）, carries on feature extraction by maximizing the ratio of MBCS to MWCS, so as to ensure that the samples in different classes are far away from each other, while the samples in the same class are as close as possible. [Result/conclusion] The comparative experiment results indicate that the method is valid.