Thou shalt not hate: Countering Online Hate Speech

Hate content in social media is ever increasing. While Facebook, Twitter,Google have attempted to take several steps to tackle this hate content, theymost often risk the violation of freedom of speech. Counterspeech, on the otherhand, provides an effective way of tackling the online hate without the loss offreedom of speech. Thus, an alternative strategy for these platforms could beto promote counterspeech as a defense against hate content. However, in orderto have a successful promotion of such counterspeech, one has to have a deepunderstanding of its dynamics in the online world. Lack of carefully curateddata largely inhibits such understanding. In this paper, we create and releasethe first ever dataset for counterspeech using comments from YouTube. The datacontains 9438 manually annotated comments where the labels indicate whether acomment is a counterspeech or not. This data allows us to perform a rigorousmeasurement study characterizing the linguistic structure of counterspeech forthe first time. This analysis results in various interesting insights such as:the counterspeech comments receive double the likes received by thenon-counterspeech comments, for certain communities majority of thenon-counterspeech comments tend to be hate speech, the different types ofcounterspeech are not all equally effective and the language choice of usersposting counterspeech is largely different from those posting non-counterspeechas revealed by a detailed psycholinguistic analysis. Finally, we build a set ofmachine learning models that are able to automatically detect counterspeech inYouTube videos with an F1-score of 0.73.