An email classifier based on resemblance

Abstract

In this paper, we study the email classification problem. We apply the notion of shingling to capture the concept of phrases. For each email, we form a sketch which is compact in size and the sketch of two emails allows for computation of their resemblance. We then apply a k-nearest neighbour algorithm to classify the emails. Experimental evaluation shows that a high degree of accuracy can be obtained.