Georgia Institute of Technology GVU Center

Since its earliest days, harassment and abuse have plagued the Internet. Recent research has focused on in-domain methods to detect abusive content and faces several challenges, most notably the need to obtain large training corpora. In this paper, we introduce a novel computational approach to address this problem called Bag of Communities (BoC)‚ technique that leverages large-scale, preexisting data from other Internet communities. We then apply BoC toward identifying abusive behavior within a major Internet community.