基于特征贡献度的特征选择方法在文本分类中应用

Application of feature selection method to text categorization based on feature contribution degree

在目前的文本分类问题中,特征选择方法被认为是提高分类精度和效率的一种有效方法.提出了一种基于特征贡献度FCD(feature contribution degree)的特征选择方法,本方法将某个特征对于类别之间区分能力的贡献度大小作为该特征被选取的条件,特征对于某一类别的FCD值为特征在该类中出现的文档数与在所有类别中出现的文档数的比值.对该方法进行了实验,并与一些常用的特征选择方法进行了比较,实验结果表明该方法具有更好的分类效果.

At present,the feature selection method is viewed as an efficient method for improving the accuracy and efficiency of classification in text categorization.A feature selection method based on feature contribution degree（FCD） is proposed.In this method,a feature will be selected according to the contribution degree for differentiating a certain category from others.The FCD of a feature in a particular category is a ratio between the number of documents occurring in a certain category and the number of documents occurring in all categories.The experiments corresponding to this method are carried out and the comparison between this method and other common feature selection method is presented.As a result,the method proposed outperforms other feature selection methods.