Abstract

Spam is a key problem in electroniccommunication. Especially in large-scale email systems.Content-based filtering is one mainstream method ofcombating this threat in its forms, an e-mail filtering systemcan learn directly from a user’s mail set, but the previousContent-based filtering methods are hard to find a balancebetween efficiency and effectiveness. Such algorithms of textcategorization as Naïve Bayes, kNN, Decision Tree andBoosting can be applied in spam filtering. However, theeffectiveness of Naïve Bayes is limited and it is not fit forinstant feedback learning. Others algorithm such as SVMare more effective but complicated to compute. Because in areal email system a large volume of emails often need to behandled in a short time, efficiency will often be as importantas effectiveness when implementing an anti-spam filteringmethod. So we intend to find a linear classifier to solve thisproblem, two online linear classifiers: the Perception andWinnow were explored for this task, which are two fastlinear classifiers. The training of these two methods is onlineand mistake driven. Furthermore, they are suitable forfeedback. We employ the two methods in three benchmarkcorpora, including PU1, Ling spam and 2005-Jun, theexperiments in public e-mail corpus show an effective result.We conclude that the two online linear classifiers have astate-of-the-art performance for filtering spam, especiallyfor Chinese spam emails.