Semantic Representation Based on Deep Learning for Spam Detection

This paper addresses the email spam filtering problem by proposing an approach based on two levels text semantic analysis. In the first level, a deep learning technique, based on Word2Vec is used to categorize emails by specific domains (e.g., health, education, finance, etc.). This enables a separate conceptual view for spams in each domain. In the second level, we extract a set of latent topics from email contents and represent them by rules to summarize the email content into compact topics discriminating spam from legitimate emails in an efficient way. The experimental study shows promising results in term of the precision of the spam detection.