基于统计频率的文本分类特征选择算法研究

A Feature Selection Method for Text Classification Based on Statistical Frequency

通过分析χ2统计量(Chi-square,CHI)的缺陷和不足,针对它对低文档频的特征项不可靠,而且不能说明词条和类别的相关性的缺点,对其进行改进,提出统计频率(Statistical Frequency,SF)算法。实验结果表明,统计频率算法能够弥补这些不足,在文本分类中表现出良好的分类效果。

This paper analyzes Chi -square algorithm （CHI） , which is unreliable for low- document frequency, and can＇t show the pertinence for term and classification. A new Statistical Frequency algorithm （SF） is proposed according to the chief shortcomings. The experiments of the SF algorithm is validated by comparison, the results show that improved algorithm performs better.