Filtering Spam at E-Mail Server Level with Improved CRM114

Abstract

This paper explores the possibility of advanced techniques at the server level, and concludes that simple improvements to a well-known Naive Bayes-based technique could turn this algorithm into a much faster and significantly better one that could be used at the server level. The problem of automatically filtering unwanted e-mail messages is one of increasing importance, because bulk mailers take advantage of the great popularity of the electronic mail communication channel for indiscriminately flooding e-mail accounts with unwanted advertisements. There are many factors that contribute to the proliferation of unsolicited spam, especially the inexpensive cost of sending e-mail,14 and of obtaining pseudonyms.15 On the other hand, we have the high cost associated with users receiving spam2 and the network overflow.