Automatic hate speech detection on social media: Turkish tweets as an example

Massive rise of user-generated web content, in particular on social media networks, caused research in social media to grow substantially in the last decade. Researches show that people may use popular micro blogging websites, especially Twitter, to make offensive comments that cover hate speech and such kind of contents can lead to a negative impact on the society. In this thesis we focus on Turkish context and investigate methods to detect hate crime incidents against ethnic/religious minorities, the LGBT population and women in social media automatically. We also evaluate the cross-domain performance of our approach and try to identify common discourses towards those groups aiming to establish a lexical baseline. Applying several supervised classification methods, we obtain out best results of 93% accuracy in detecting sexist tweets. Other results demonstrate that the main challenge lies in discriminating racist tweets.