Word Representation Learning for Detecting Malicious Chat Messages

Communicating with peers online is an increasingly popular activity for children and teenagers, which has led to growing concerns about bullying and other forms of malicious behaviour in online chat rooms and virtual worlds. Although some online forums are patrolled by human moderators, the amount of text being generated is typically too large even for a dedicated team of humans to process. This research project will adapt machine learning techniques for the task of automatically detecting and filtering malicious messages in online chat rooms and virtual worlds. The partner organization, Two Hat Security, provides custom chat filtering as a web service, with the goal of keeping kids safe online. The expected benefits of this internship to Two Hat Security are an improvement in the accuracy of its chat filtering system, as well as more efficient and effective customization of filters for the different virtual worlds that it monitors on behalf of its clients.

Faculty Supervisor:

Dr. Dale Schuurmans

Student:

Kenneth Dwyer

Partner:

Two Hat Security Research Corp.

Discipline:

Computer science

Sector:

Information and communications technologies

University:

University of Alberta