Identifying spam without peeking at the contents

Abstract

Currently available solutions attempt to filter out spam based on analyzing the contents of the message and calculating a score to indicate the 'spami-ness' of the message. However, users can typically identify their junk email without having to open and read the contents of the specific message. In this article, we outline the general problem, review current options, and propose a new user-level behavior model to identify spam messages. We show the performance of this approach and discuss some applications and future directions.