CORAL: Uncertainty-Aware Regulation of Exposure Concentration in Recommender Systems

Recommender systems (RS) may suffer from feedback-driven exposure concentration, where repeated engagement optimization collapses exposure onto a narrow set of categories, reducing catalog coverage and degrading long-horizon learning. Existing methods are often post hoc and typically lack principled uncertainty-aware risk estimates for regulating exposure under endogenous feedback. We therefore propose CORAL, a model-agnostic, uncertainty-aware framework that formulates exposure regulation as a constrained sequential decision problem. Specifically, we model self-reinforcing interactions to construct an exposure-saturation state, then derive an upper confidence bound on category-conditioned violation risk from observed history and incorporate it through a state-dependent penalty for adaptive intervention near saturation. Moreover, we provide theoretical guarantees for risk bounds, finite-time recovery, and efficient long-term performance. Extensive experiments on real-world datasets and controlled simulations validate the effectiveness of the proposed framework, which aligns with our theoretical analysis. Our code is available at: https://github.com/downw/CORAL.

Recommendation systems often learn from users’ past clicks and then use those clicks to make future recommendations. Over time, this feedback loop can cause the system to recommend too many items from the same category, creating an echo-chamber effect and reducing users’ exposure to other useful content. This paper introduces CORAL, a method that monitors when a user’s recommendations are becoming overly concentrated and adjusts the recommendation process before the system becomes too narrow. CORAL does this by estimating a user’s exposure saturation level and applying stronger regulation only when the risk of over-concentration becomes high. Experiments show that CORAL can reduce exposure concentration while maintaining strong recommendation utility across several datasets and simulated long-term interaction settings.