Abstract: Feature selection is the key stage in the text categorization,and the processing of it will affect the speed and accuracy of text classification．χ²statistic is a important methods in feature selection of text categorization since it mea-sures the dependence between a term and a class effectively．Nevertheless,we found the feature in the vectors of CHI can not fully express the means of concept and it depends the training text set,and the vectors of CHI are used only for the phase of feature selection after the analysis of the application of χ² statistic in the text categorization．So this paper proposed an improved kNN text categorization algorithm based on Semantic-Vector-Combination and Multi-class of feature,in which the feature considers the means of concept,and the matrix of multiclass of features will improve the efficiency of algorithm in the stage of categorization．The results and analysis of experiments show that the efficiency of categorization is improved and its accuracy is also enhanced.

Key words: Text categorization,χ²statistic,Feature-MultiClass-Matrix,K-Nearest neighbor