Aggressive Text Detection for Cyberbullying

Aggressive text detection in social networks allows to identify offenses and misbehavior, and leverages tasks such as cyberbullying detection. We propose to automatically map a document with an aggressiveness score (thus treating aggressive text detection as a regression problem) and explore different approaches for this purpose. These include lexiconbased, supervised, fuzzy, and statistical approaches. We test the different methods over a dataset extracted from Twitter and compare them against human evaluation. Our results favor approaches that consider several features (particularly the presence of swear or profane words).