aiqbench analysis
This study empirically demonstrates that X's recommender system inadvertently encodes user ideology within its embedding space, effectively profiling political positions as a by-product of relevance optimization. The research reveals that this specific dimension is distinct from demographic factors like age or gender and significantly influences the diversity of recommended accounts. By removing this ideological scale, the authors achieved greater recommendation diversity with only a minor trade-off in accuracy, highlighting a critical gap between current legal definitions of profiling and emergent algorithmic behaviors.
Recommender system in X inadvertently profiles ideological positions of users
arXiv:2602.02624v2 Announce Type: replace-cross Abstract: Several data protection laws restrict processing that reveals political opinions, irrespective of the controller's intent. Whether recommender systems do so as a by-product of optimizing relevance has not been measured. From 2.5 million ``Who to Follow'' recommendations shown to 682 volunteers in France, we reconstructed an approximation of the embedding used by X's recommender for 26,509 accounts, computing survey-calibrated ideology scores. One direction in this embedding orders users by Left-Right position (Pearson rho = 0.887), distinct from directions tracking age, gender or popularity. We show this scale exists and affects the recommendations computed from the embedding. Removing it diversified recommendations at a limited cost in accuracy. We document a consequential form of emergent political representation that current definitions of profiling do not clearly address.
Abstract:Several data protection laws restrict processing that reveals political opinions, irrespective of the controller's intent. Whether recommender systems do so as a by-product of optimizing relevance has not been measured. From 2.5 million ``Who to Follow'' recommendations shown to 682 volunteers in France, we reconstructed an approximation of the embedding used by X's recommender for 26,509 accounts, computing survey-calibrated ideology scores. One direction in this embedding orders users by Left-Right position (Pearson rho = 0.887), distinct from directions tracking age, gender or popularity. We show this scale exists and affects the recommendations computed from the embedding. Removing it diversified recommendations at a limited cost in accuracy. We document a consequential form of emergent political representation that current definitions of profiling do not clearly address.
Social and Information Networks (cs.SI); Artificial Intelligence (cs.AI); Computers and Society (cs.CY)
arXiv:2602.02624 [cs.SI]
(or arXiv:2602.02624v2 [cs.SI] for this version)
https://doi.org/10.48550/arXiv.2602.02624 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 225 words
Related coverage
State of the Field: AI for Epistemics and Coordination
LessWrong · Oct 2, 2026
Lock-In Risk Needs More Researchers; Here's Where to Start
LessWrong · Jun 17, 2026
Textual User Taste: Natural-Language User Context for Foundation-Model Recommender System at Scale
arXiv · Sep 29, 2026
[AINews] Reality Checks on AI News (Yegge shuts down Gas Town, Databricks’ +60% Astra cost)
Latent Space · Sep 17, 2026