Abstract:
A growing body of research focuses on computationally detecting controversial topics and understanding the stances people hold on them. Yet gaps remain in our theoretical and practical understanding of how to define controversy, how it manifests, and how to measure it. Since controversy is a complicated social phenomenon, it is difficult to understand what elements make up the controversy. Previous work has attempted to capture controversy algorithmically by studying cues for disagreement and polarity between different stance groups, but lacked a systematic understanding of how controversy should be defined and measured. In this work, I propose a multi-dimensional model of controversy. Specifically, this model demonstrates that: (a) controversy is not a fixed universal quantity, but rather is highly dependent on the population observed; and (b) controversy is not a unidimensional entity, but rather comprises at least two dimensions, contention and importance. I model contention and importance within a population from a mathematical standpoint. To validate and evaluate the soundness of our theoretical model, I instantiate the model to algorithms for a diverse set of sources: polling, Twitter, and Wikipedia. I will demonstrate that our controversy model holds an explanatory power for observed phenomena but also a predictive power for tasks, such as identifying controversial Wikipedia articles. Additionally, I will briefly present recent analysis, using this model and the contention score, on a proprietary dataset of Parler posts and comments from January 6-11, prior to its shutdown. I will also discuss plans for future research.
