基于类别的混合式特征选择方法在文本分类中的研究

The research of text categorization based on the hybrid feature selection method

针对传统的最大特征选择方法(CMFD)存在类间分布不均匀与类内个别样本分布稀疏从而导致分类准确度下降的问题,提出一种基于类别的混合式最大特征选择方法(HCMFD)。该方法首先在预过滤阶段中引入相关度,冗余度与协调度的概念提取出具有最大特征区分度的最小子集,然后在后过滤阶段中对每个类别赋予不同的阈值,通过把类别阈值和特征评估函数(FEF)得到的最佳匹配度进行比较得到最终的特征选择集合。实验证明改进的特征选择方法(HCMFD)比原有的方法(CMFD)具有较高的分类准确度和更高的预测水平。

According to the characteristics of the traditional maximum selection method(CMFD)is unevenly distributed among and within individual samples which leads to sparse distribution of classification accuracy decline,proposes a hybrid feature selection method based on the maximum category(HCMFD).Firstly,in the pre filtration stage in the introduction of the concept of relevance and redundancy of the features extracted with the maximum discrimination of the minimal set of,The final set of feature selection is obtained by comparing the category matching and the best matching obtained by the feature evaluation function(FEF)The experimental results show that the improved feature selection method(HCMFD)has higher classification accuracy and higher prediction level than the original method(CMFD).