CS547 Human-Computer Interaction Seminar (Seminar on People, Computers, and Design)
Fridays 11:30am-12:30pm PT · Gates B3 · Open to the public|
Magdalena Wojcieszak
UC Davis
Interest-based and Ideological Biases in Online Content Consumption: Problems and Solutions
April 11, 2025
Populism, polarization, misinformation, and wavering support for democratic norms are pressing threats to many democracies. Although the sources of these threats are multifaceted, social media platforms and their recommendation algorithms are often seen as the culprit. Many observers and scholars worry that platform algorithms create filter bubbles and lead to rabbit holes of radicalization. In this presentation, I briefly mention four projects that address these questions by merging computer science tools and methodologies with large-scale human subjects work. First, I present a sock puppet-based audit of ideological biases on YouTube, which allows us to systematically and at scale isolate the influence of the algorithm in recommendations to congenial and radical content (Haroon et al., 2023, PNAS). Second, I argue that although most scholars worry about ideological biases in recommender systems, interest-based biases are as -- if not more -- democratically consequential. The problem is less that people consume "bad" political content (radical, unverified, or otherwise problematic), but that most do not consume any at all (Wojcieszak et al., 2023; IJPP). I present two computational interventions, a field experiment on Twitter and an over-time experiment using a browser extension for YouTube, which aimed to incentivize citizen exposure to quality and diverse news and political content. |
|