Abstract

Online social platforms use recommender algorithms to collate and sort the universe of messages users see, which is distorted in how popular content will be perceived before any personalization. We call this “exposure bias,” and we focus on evaluating different recommendation and personalization approaches using diverse exposure bias metrics to understand how ranked content affects user perception. Similarly we simulate user activity in a social network to assess the influence of such ranking approaches on exposure bias. Furthermore, we are working on parsimonious agent-based models to explore intervention effects over time. Our empirical findings reveal that users exposed to personalized feeds from various recommender system types experience more exposure bias than reverse chronological and popularity-based feeds.

Metadata

publication

Social Network Analysis and Mining Applications in Healthcare and Anomaly …, 2024

year

2024

publication date

2024/10/9

authors

Nathan Bartley, Keith Burghardt, Kristina Lerman

link

https://link.springer.com/chapter/10.1007/978-3-031-75204-9_13

book

Social Network Analysis and Mining Applications in Healthcare and Anomaly Detection

pages

317-336

publisher

Springer Nature Switzerland