Catching the picospams

Abstract

In this paper, we study the problem of filtering unsolicited bulk emails, also known as spam emails. We apply a k-NN algorithm with a similarity measure called resemblance and compare it with the naive Bayes and the k-NN algorithm with TF-IDF weighting. Experimental evaluation shows that our method produces the lowest-cost results under different cost models of classification. Compared with TF-IDF weighting, our method is more practical in a dynamic environment. Also, our method successfully catches a notorious class of spams called picospams. We believe that it will be a useful member in a hybrid classifier. © Springer-Verlag Berlin Heidelberg 2005.