Paul Anton Bachmann, Hannah Eckert, Oleg Lesota, Martin Lackner, Markus Schedl · ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 2026
DOI: 10.1145/3799682.3840023
Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).
Ensuring fair exposure across demographic groups of content creators in recommender systems is crucial, as it affects the real-world visibility of minorities and underrepresented groups. Existing work on provider-side exposure fairness has primarily considered single demographic attributes, leaving multi-attribute and intersectional exposure fairness largely unexplored. Yet in real-world platforms, fairness concerns may arise simultaneously across multiple intersecting provider attributes. We introduce multi-attribute and intersectional perspectives on exposure fairness in recommender systems, propose an evaluation framework taking both perspectives into account, and present a post-processing approach for multi-attribute exposure fairness informed by online settings in social choice theory. Our experiments on real-world and synthetic datasets show a clear distinction between the two perspectives, as baseline approaches demonstrate a tendency to trade attribute-level fairness for improvements on the intersectional level. Our proposed method achieves a more favorable fairness-utility trade-off across attribute-wise exposure objectives and robust improvement of both exposure fairness aspects on the real world data. On synthetic data no method shows consistent improvement of both attribute-level and intersectional exposure fairness across all backbone models, highlighting the challenging nature of the task and the need to evaluate multi-attribute exposure fairness from both perspectives. With this work we aim to kick off the research on provider exposure fairness in recommender systems at a more granular level.
No comments yet — start the discussion below.