Email User Classification and Topic Modeling

Abstract

Nowadays, the primary form of communication in most organizations is via email. Thousands of users communicate daily via email. And with an increasing amount of emails comes the fraudulent activities. It is necessary to identify the righteous user who wrote the email in any organization to avoid misuse of the mail server. Also, from a pool of emails, it can be helpful to filter out emails that belong to keywords of interest to focus on a certain topic and find out relevant users during any investigation or research. This paper proposes a method to analyze the content of the email and converts the textual data into vectorized features to predict the author of the email based on writing patterns of different users using Deep Learning methods. Along with that, this paper also leverages existing methods such as DBSCAN and LDA along with their performance to cluster similar emails with each other for the extraction of topics and keywords from the contents of emails. These techniques can serve as a beneficiary tool for any investigating party to learn important features and extract maximum information from the email corpus.