Aiming at problem that traditional CHI method chooses the feature in the global scope and ignores the information of word frequency

an improved feature selection method is proposed. By introducing feature distribution correlation coefficient

localized strong correlated features are selected

and the correction factor is used to solve the negative correlation obsession of the CHI method

thus the classification index of corpus set is improved. When experiments are conducted on Netease news corpus and Fudan University Chinese corpus

feature selection is first performed by using the above method

and then the selected features are weighted by the improved term frequency-inverse document frequency（ TF-IDF） weighting formula. The classifier uses support vector machine（ SVM） and naive Bayes method. The experimental results show that the improved algorithm not only shows a significant improvement in classification effect

but also has more stable performance.