Robust Detection of Cyberbullying in Social Media

The potentially detrimental effects of cyberbullying have led to the development of numerous automated, data-driven approaches, with an emphasis on classification accuracy. Cyberbullying, as a form of abusive online behavior, although not well-defined, is a repetitive process, i.e., a sequence of aggressive messages sent from a bully to a victim over a period of time with the intent to harm the victim. Existing work has focused on aggression (i.e., using profanity to classify toxic comments independently) as an indicator of cyberbullying, disregarding the repetitive nature of this harassing process. However, raising a cyberbullying alert immediately after an aggressive comment is detected can lead to a high number of false positives. At the same time, three key practical challenges remain unaddressed: (i) detection timeliness, which is necessary to support victims as early as possible, (ii) scalability to the staggering rates at which content is generated in online social networks, (iii) reliance on high quality annotations from human experts for training of highly accurate supervised classifiers. To overcome the challenges associated with cyberbullying detection in online social networks, my PhD thesis focuses on a novel formulation of the online classification problem as sequential hypothesis testing that seeks to drastically reduce the number of features used while maintaining high classification accuracy. To reduce the dependency on labeled datasets, I seek to develop efficient semisupervised methods that extrapolate from a small seed set of expert annotations. Preliminary results are very encouraging, showing significant improvements over the state-of-the-art.