Balancing Relevance and Responsibility: A Multi-Stakeholder Equilibrium Framework for Recommendation Systems

Traditional recommender systems predominantly optimize for a single objective: maximizing user relevance or engagement. This greedy optimization often leads to negative externalities, including the proliferation of misinformation, polarization, and unfair exposure for content creators. To address these issues, we propose a Multi-Stakeholder Equilibrium Framework that formulates recommendation as a Social Welfare Maximization problem. Our approach explicitly models the utilities of four distinct stakeholders: Users, Creators, the Platform, and Society. Using a modified MovieLens dataset infused with synthetic controversy scores, we benchmark our equilibrium approach against a standard relevance-maximizing greedy baseline. The results demonstrate that the equilibrium framework significantly reduces societal risk—lowering misinformation and polarization by approximately 28%—while increasing content diversity by 15.7%. Notably, these improvements are achieved with only a marginal decrease (9.5%) in user relevance. This study provides empirical evidence that constrained multi-objective optimization can effectively balance profitability with social responsibility in recommendation ecosystems.