Applications of Supervised and Unsupervised Ensemble Methods

Abstract

In their daily work and common life, people suffer serious problems with Unsolicited Commercial E-mails (UCE), commonly known as spam : common people, small companies and large public or private institutions feel that spam has weakened the reliability and effectiveness of email as an efficient tool for communicating. To establish simple, fast and effective countermeasures against spam attacks is a necessary strategy of a modern mailing management system. In this chapter we describe a novel method for detecting spam messages, analyzing both text and image attached components. In particular, we describe an architecture for deploying a personal antispam system able to overcome some problems that are still besetting the state-of-the-art spam filters. Text analysis is accomplished by considering recent advances in both semantic and syntactic analysis; in addition, spammers tricks based on images are also taken into account. A Behaviour Knowledge Space approach for fusing the different results coming from the analysis of the different parts of the e-mails enhances the performance of the proposed system, as described by the experiments we have carried out. © 2009 Springer-Verlag Berlin Heidelberg.