The Power of Diffusion Networks: Learning to Detect the Spread of Hate Speech Online

Abstract

Researchers have yet to fully understand the online spread of divisive content. As hate groups continuously seek to avoid censorship, building new detection methods is an important countermeasure. This thesis proposes a framework for detecting hate speech messages on Twitter based solely on the network topology and time dynamics of their spread. I produce network-based classifiers identifying hate speech with sufficient accuracy and recall, relying on an application of kernel methods to social network analysis. My classification techniques identify instances of hate speech missed by purely text-based classification models, suggesting that network-based methods can complement existing automated hate speech detection systems.