Feature selection is a vital preprocessing step for text classification task used to solve the curse of dimensionality problem. Most existing metrics (such as information gain) only evaluate features individually but completely ignore the redundancy between them. This can decrease the overall discri