A Method of Text Feature Extraction Based on Weighted Scatter Difference

摘要

Feature reduction is one of the core technologies of automatic text categorization. As for the scatter difference criterion, poor categorization effect is made when the between-class distance is small and the class density is high. In order to solve this problem, a weighted method based on the sample distribution is shown in the paper, which will make the between-class and within-class scatter matrixes with poor scatter be weighted, to enhance the categorization ability after dimensional reduction and to improve the dimensional reduction effect of linear feature extraction method based on scatter difference. The following experiment tells us that this method is superior to the original maximum scatter difference method in precision rate and recall rate.