Revisit Bayesian Approaches for Spam Detection

Due to fast changing of spam techniques, we argue that multiple spam detection strategies should be developed to effectively against spam. Among many others, (naive) Bayesian filter is one of those being proposed. While there are some early works being done on Bayesian approaches for spam detection, it is unknown if such a simple approach is still effective to filter spam mail today. In this paper, we re-evaluate the Bayesian approach with a public spam database. We found that Bayesian approach can achieve high spam detection rate for plain-text mail. However we found, at the same time, some practical issues to use Bayesian filters, for example multimedia contents and different message encoding schemes for non-English characters, are needed to be carefully handled.