Detecting age in online chat

Abstract

Over 90% of teens in the United States use the Internet, and many use it for social interaction. Due to the faceless nature of digital communication, criminals can easily pose as legitimate users to build friendship and trust with potential victims. Even though fewer youths are going to chat rooms and talking to people they do not know, the number of youths receiving aggressive solicitations for offline contact has not declined. Most sexual solicitations go unreported to law enforcement and parents. Though it is a crime for an adult to sexually exploit a minor, it is not always a crime for teens to solicit other teens. It would be of great help to law enforcement agencies if they could automatically detect adults soliciting teens versus teens soliciting other teens in online chat. This study analyzes the effectiveness of different machine learning techniques to distinguish chat conversation by teens and adults. Using proposed techniques, we classified teen and adult conversations with an accuracy of 86%. The goal of this research is to build an automatic recognition system of adults conversing with teens, capable of detecting predators and alerting agencies or parents of possible inappropriate conversations.