Quantifying Toxicity and Verbal Violence on Twitter

Online harassment is a continuing problem, endemic to many social media platforms and other means of web-based communications, and few means exist to analyze web content for instances of verbal violence and aggression. We are developing a scale of online aggression that can be applied to Twitter posts (tweets) and that is based on existing measures of trait aggression and cyberbullying. For the purpose of testing and validating our scale, we are relying on Mechanical Turk, an Amazon Web Service, through which we can enlist and pay workers to code our dataset of tweets. Preliminary results suggest that aggression in tweets is difficult for human coders to identify and that we have not reached consensus about what constitutes harassment online. We discuss our preliminary results and propose next steps such as scale modification and automated classifier development.