Multi-level filters for a Web-based e-mail system

With spams becoming an ever increasing problem, different strategies have been used to fight against these junk e-mails, including feature-based and statistic approaches. Although these approaches are effective in filtering out the junk e-mails, they frequently generate false positives. In this paper, we describe a multilevel cascading filter approach for a Web-based email system. The system allows the deployment of any combination of three filtering levels: a cascade of individual filters, a rule-based filter, and a naive Bayesian filter. The first level of individual filters reduces the rate of false positives and false negatives, while the latter two levels increase the efficiency of the filter system. Preliminary tests on a corpus of spam and nonspam e-mails yield some satisfactory results. Additionally, a set of e-mail delivery agents is also implemented in the Web email system.