Speaker: Cody Buntain, Assistant Professor, iSchool, UMD


Location: HBK 2105

Abstract: While originally developed to increase diversity in product recommendations and show individuals personalized content, recommendation systems have increasingly been criticized for their opacity, potential to radicalize vulnerable users, and incentivizing anti-social content. At the same time, studies have shown that modified recommendation systems can suppress anti-social content across the information ecosystem, and platforms are increasingly relying on such modifications for soft content-moderation interventions. These contradictions are difficult to reconcile as the underlying recommendation systems are often dynamic and commercially sensitive, making academic research on them difficult. This paper sheds light on these issues in the context of political news consumption by building several recommendation systems from first principles, populated with real-world engagement data from Twitter and Reddit. Using domain-level ideology measures, we simulate individuals’ ideological trajectories through recommendations for news sources and examine whether standard recommendation approaches drive individuals to more partisan content and under what circumstances such radicalizing trajectories may emerge. We end with a discussion of personalization’s impact in consuming political content, and implications for instrumenting deployed recommendation systems for anti-social effects.

Bio: Dr. Cody Buntain is an assistant professor in the College of Information Studies at the University of Maryland and a research affiliate for NYU’s Center for Social Media and Politics, where he studies online information and social media. His work examines how people use online information spaces during crises and political unrest, with a focus on information quality, preventing manipulation, and enhancing resilience. His work in these areas has been covered by the New York Times, Washington Post, WIRED, and others. Prior to UMD, he was an assistant professor at the New Jersey Institute of Technology and a fellow at the Intelligence Community Postdoctoral Fellowship.