Recommender Systems as Control Systems
Provides a new theoretical framework for understanding long-term dynamics of fairness in recommender systems, relevant to researchers and practitioners designing fair algorithms.
The paper proposes a control-theoretic interpretation of recommender systems and shows that fairness interventions, when optimized over time, can improve overall system performance rather than simply trading off against utility.
We propose a control-theoretic interpretation of recommender systems and use this perspective to analyze how fairness interventions shape long-term system behavior. Fairness concerns arise for both users and creators, ranging from opinion polarization and representation bias on the user side to popularity bias on the creator side. A central insight of our analysis is that fairness should not be viewed as a simple trade-off against utility. When optimized over time, it can in fact be beneficial for overall system performance. Realizing these gains, however, requires a clear understanding of the underlying dynamics.