Skip to Main Content
Internet security is seriously threatened by spam spreading, and content-based spam filtering has become one of effective spam-filtering methods. Aiming at the practical problems, we propose an active learning based method which takes naive Bayesian means as basic classifiers. This method randomly initialize a small training set to generate basic classifiers, and then use them to classify mails, which add the most uncertain mail to training set each time to improve the classifier performance. The simulations based on the CCERT mail set show that this method not only reduces the number of mails to be labeled, but also improves classifier accuracy.
Date of Conference: 7-9 July 2010