A Novel Approach for Email Clustering Based on Semantics

Abstract

An increasing interest has been recently devoted to clustering short documents. Short documents don't contain enough text to compute similarities accurately by implementing the most widely used technique called Vector Space Model (VSM). Adding semantics to short documents clustering is one efficient way to solve this problem. However, real life collections are often composed of very short or long documents. For example, the length of email messages for each email user follows a power-law distribution. Long emails and short emails both appear in email corpus. Therefore, both state-of-the-art short documents and long document clustering approaches can't get a high cluster quality or high efficiency in short and long documents clustering. In order to solve this problem, we propose a novel approach for email clustering based on semantics. Empirical validation shows that our method can obtain high cluster quality and high efficiency in real world email datasets.