About the talk

Over the last decades, there have been increased concerns over deviant behaviors in online communities. Detecting and classifying these behaviors is a challenging task for moderators and automated systems since there does not exist a unique definition of each behavior. Consequently, the moderation process can be subjective and biased.

To this end, drawing from a large literature review of theoretical, empirical, and practical studies, we propose in this work a standardized taxonomy of online deviant behaviors that unifies all available definitions. We identified seven main categories of deviant behaviors: cyberbullying, sexting, online sexual victimization, self-harm, suicide, online hate and swearing. Each category groups different subcategories of deviant behaviors.

To validate the developed taxonomy, we computed an inter-annotator agreement. We expect that this standardized set of definitions will reduce or eliminate the ambiguity that exists in labelling deviant behaviors, and will reduce subjectivity in moderation rules.

Speaker : Zeineb Trabelsi, Ph.D. candidate in Administration Science, Université Laval