Abstract: On the basis of multiple hypothesis testing, we proposed a feature weighted naive Bayesian algorithm, which outputs many sets of feature words by means of feature selection, and assigns a coefficient to each set of feature words which is used to construct the classifier in terms of the error rate of multiple hypothesis testing. This algorithm was used in the text classification of the mayor’s public access line project, where we realized the automatic classification of complaint texts by constructing three feature weighted naive Bayesian classifiers. Compared with those of the traditional methods, the efficiency and accuracy of our classifier are higher.

Key words: multiple hypothesis testing, text classification, fea ture weighted, the mayor’s public access line project