PRIS Kidult Anti-SPAM Solution at the TREC 2005 Spam Track: Improving the Performance of Naive Bayes for Spam Detection

Recently, the spam already constituted a serious problem for both e-mail users and Internet Service Providers (ISP). Solutions to the abuse of spam would be both technical and legal regulatory. This paper reports our solution for the TREC 2005 spam track, in which we consider the use of Naive Bayes spam filter for its desirable properties (simplicity, low time and memory requirements, etc.). Then the approaches to modify the Naive Bayes by simply introducing weight and classifier assemble based on dynamic threshold are proposed, which can help to improve the accuracy of a Naive Bayes spam classifier dramatically. Additionally, we discuss some steps that must be adopted naturally thought before, such as stop list, word stemming, feature selection, class prior probabilities. The theory analysis implies these steps are not necessarily the best way to extend the Bayesian classifier, and these were also verified empirically. Many of these techniques appear to be counterintuitive but can be explained by the statistical properties of e-mail itself. Experiment results of TREC 2005 spam track demonstrate the effectiveness of the proposed method.