Social media platforms have enabled the vast proliferation of online hate messages and the growth of online hate groups worldwide. Considerable efforts in the computational sciences have thus been dedicated to detecting online hate. However, common approaches are constrained by their binary classification of individual texts. Binary classification schemes collapse the unique social narratives of hate used to target various identities. Focusing on individual texts likewise erases the social networks through which hate spreads and organizes communities. Such shortcomings spill over into downstream challenges, such as tracking the manipulation of online hate or identifying its causes and effects in the offline world. Without attending to narratives and networks, these assessments are flattened into piecemeal questions of how much hate is present or which individuals are spreading it. Consequently, interventions to mitigate online hate fail to account for its deeply social character. Hence, this thesis asks: How can we characterize the narratives and networks of online hate? How can narratives and networks of online hate help us better understand its influences and impacts?

This thesis integrates social scientific theory and computational methods to address these questions. First, I use social identity theory to reframe online hate from a decontextualized linguistic phenomenon into a product of group conflict in society. Social identities set the stage for viewing online hate in terms of its wider social narratives and networks, and motivate a theory-based model to detect online hate that is faster, more generalizable, and more explainable than existing methods. Second, I introduce computational social science tools for characterizing online hate narratives and networks. I showcase the diverse semantic features of online hate narratives targeting racial, gendered, political, and religious identities. I also capture the structural features of online hate networks in terms of properties such as density, isolation, and hierarchy. Third, I develop novel measures for the manipulation of online hate narratives and networks. These indicators quantify shifts in the content of online hate narratives and the organization of online hate networks due to the activity of bots and trolls. Finally, I connect online hate in social media conversations with offline contexts of societal upheaval. I empirically investigate how variations in offline conditions trigger changes in online hate, which in turn predicts offline violence. Collectively, these contributions enhance our social scientific understanding of the nature, influences, and impacts of online hate, and underscore the importance of theoretically informed computational methods to meaningfully engage it.

165 pages

Thesis Committee:
Kathleen M. Carley (Chair)
Patrick Park
Daniel M. Oppenheimer
James Hawdon (Virginia Institute of Technology)

James D. Herbsleb, Head, Software and Societal Systems Department
Martial Hebert, Dean, School of Computer Science

Return to: SCS Technical Report Collection
School of Computer Science

This page maintained by reports@cs.cmu.edu