PLEASE NOTE DATE/TIME/ROOM CHANGE FOR MAY'S DEFENSE!
Since 2014, there has been a steady rise in the number of online hate groups in the U.S.; hate groups are now present in each of the 50 states for the very first time in eight years. Given the wide reach of social media, many hate groups leverage social networks to not only propagate hate messages but also grow their base. In this talk, I present the first linguistic analysis of the dynamics of the most prevalent hate ideologies in the U.S. based on their Twitter footprints. I discuss drives that control hate ideologies discourse and I present their semantic similarity evolution over the course of 2015-2017. I then present implications of our results for next-generation hate speech detection systems. Finally, I show that stance detection in the context of hate speech is vital. Lexicons of hate speech keywords neglect the stance of the speaker. I present a framework of four stances: directed hate, generalized hate, constructive counter hate, and benign content. I present the first dataset that captures these four stances for different hate speech keywords. I conclude by discussing the challenges related to labeling content related to hate speech and how this dataset could be used to improve role labeling systems in the context of online abuse.