Abstract
Conspiracy theory belief and other uncivil expressions are on the rise in online spaces. The social media companies in charge of curating these spaces have been reluctant to address these issues in ways that oppose their commercial interest. This reality has prompted the research community to study the nature of online conspiracy theories and the policies that could control them. Unfortunately, the black-box nature of social media recommender systems has distorted the types of research questions that can be asked of the platforms. Social science studies focus on the traits of people who are vulnerable to extremism, while computational studies skip ahead to automated content moderation schemes.
This dissertation provides a bridge between these two research topics. Most social media sites are organized as social network graphs, which can be examined using graph algorithms. We use graph centrality and network flow to answer the theoretical questions of why certain users become embedded in online conspiracy communities and what tactics are most effective at introducing new users. We then demonstrate two novel schemes for examining real-world social network data and determining which topological factors of the network are most conducive to changing users’ opinions and potentially motivating their actions. These analyses find that smaller, denser groups that mimic an offline social circle are the most effective at increasing ideological homogeneity in an information flow network and at motivating action in a social network. These results provide information that can be used by industry leaders, public advocates, and even individual users to mitigate the effects uncivil expressions online have on social media sites.