Workshop Proceedings of the 18th International AAAI Conference on Web and Social Media

Workshop: CySoc 2024: 5th International Workshop on Cyber Social Threats

DOI: 10.36190/2024.12

Published: 2024-06-01
Auditing Exposure Bias on Social Media for a Healthier Online Discourse
Nathan Bartley, Keith Burghardt, Kristina Lerman

Just as they are responsible for curating what users see in their personalized feeds, social media platforms also curate who you see. This is not a common object of study in platform audit studies, as the unit of analysis tends to be found in the content. With approximately 24 sock puppet accounts on X (formerly known as Twitter), we show that biased simple interactions skew a users' perception of their local network when compared to random simple interactions. We further show that reverse chronological and personalized timelines behave differently, with personalized timelines showing significantly underestimated proportion of pro-science content than reverse chronological timelines. This suggests that personalization mixes with end-user activity to expose users to significantly skewed groups of friends than similar users using reverse chronological timelines. This has implications for the analysis of social media with regard to both the spread of harmful narratives and how social media may impact individuals' mental health.