Citadel: An Automated Abuse Detection System to Detect And Prevent Abusive Behaviors over Emails

Abstract

Even though emails are identified as a prominent source of exchanging abusive behaviors, very little work has explored abuse over emails. In our accepted paper in NSysS 2021, we explore perceptions of users on types of abuse detection systems for emails, revealing privacy concerns and lack of control in human-moderator-based systems and a noteworthy demand for an automated system. Motivated by the findings, we iteratively develop an automated abuse detection system "Citadel" for emails in two sequential phases and evaluate in both phases - first over 39 participants through in-person demonstrations, and second over 21 participants through a 3-day field study and over 63 participants through a video demonstration. Evaluation results portray efficacy, efficiency, and user acceptance of "Citadel" in detecting and preventing abusive emails.