Exploring Automatic Identification of Fantasy-Driven and Contact-Driven Sexual Solicitors

In an increasingly computer-mediated world, it is easier to start conversations with people online, making it convenient to build trust with people based on their online persona. This encourages people with malevolent intentions to take advantage of trustful youngsters by means of various internet-based messaging platforms and subsequently inflicting harm to people below the legal age if a physical meeting is arranged. While previous research has prioritized the automatic identification of online solicitors by means of analyzing online chat conversations, little work has been done in terms of studying the conversational features of chats involving fantasy-driven and contact-driven online solicitors. This study focuses on 271 conversations extracted from Perverted-Justice, an online repository of chats between a predator and a volunteer pretending to be underage and presents results from training a model to triage between fantasy and contact-driven online solicitors.

I. Introduction

In 2013, Sweetie, a virtual agent who mimicked the appearance of a 10 year old girl, was used by researchers to communicate with individuals interested in webcam child-sex [1]. While Sweetie was only live for 10 weeks, over 1,000 predators provided enough information to Sweetie to be personally identified, and the information was passed on to Interpol. Though the Sweetie sting was considered a success, the agent still required a human to guide the conversation with the solicitor behind the scene. An autonomous multi-agent system that incorporates behavioral modeling, biometrics, and image analysis as inputs can greatly aid the movement towards a safer Internet experience. Such a tool will also help law enforcement who struggle to keep up with the large case loads of sexual solicitation [2].