Improved Feature Selection Method Based on CHI for Text Categorization

Because the traditional Chi-square methodchooses the feature in the global scope and ignores theinformationof word frequency and distribution, this paper proposes animproved feature selection method. The method selects anumber of strong featureswith thedefined feature distributioncoefficient, andtakes into account featuredistribution thatimproves theperformances of Chi-square method in theunbalanced dataset.The results of the experimentshow that the improved algorithmnot only shows a significant improvement in classification efficiency, but also has more stable performance.