Study of E-mail Filtering Based on Mutual Information Text Feature Selection Method

摘要

Aiming at the problem with filtering E-mail, based on analyzing defects of the traditional mutual information , an approach based on quadratic TF * IDF mutual information feature selection is presented in the paper; then the importance of characteristic words appearing just in only one class is again measured to solve the problem that feature selection is not effectively done because of equal mutual information value. Finally, Bayesian classifier is used for experiment and experimental result shows that compared with the original method, the presented approach possesses higher correct rate and more efficiency of classification in text classification.