研究推荐系统中公平性与商业目标的权衡，发现多物品推荐下公平性不再免费。
Equity by Design? On the Trade-Offs in Fairness-Driven Recommendation in Heterogeneous Two-Sided Markets
- 构建统一框架，联合优化生产者曝光、用户群公平性与业务约束
- 多物品推荐下公平性需付出代价，适度公平约束可提升业务指标
- 降低群体差异不影响整体体验，适合平台设计与政策制定者
双边市场存在激励异质性：生产者追求曝光，消费者追求相关性，通过约束优化平衡二者已成为标准做法。然而实际平台面临多重异质性交互，常被孤立研究：多物品推荐、异质用户群体及超出原始相关性的业务约束。本文提出并研究一种离线优化框架，统一分析这些权衡，扩展了先前的双边模型以更真实地建模离散多物品推荐。该框架将生产者侧曝光保障与用户群体公平性目标、明确的业务约束相结合。实验表明，在每个用户接收多个推荐后，此前在简化单物品场景中报告的“免费公平性”现象消失；适度的生产者公平性约束可通过分散对饱和生产者的曝光，提升模拟业务指标。进一步发现，减少群体间差异的同时仍能保持整体竞争力的实用性。
原文摘要 · Abstract (English)
Two-sided marketplaces embody heterogeneity in incentives: producers seek exposure while consumers seek relevance, and balancing these competing objectives through constrained optimization is now a standard practice. Yet practical platforms face interacting sources of heterogeneity that are often studied separately: multi-item recommendation, heterogeneous consumer groups, and business constraints beyond raw relevance. In this work, we present and study offline optimization framework for analyzing these trade-offs in an unified manner, extending prior two-sided formulations to represent more realistic discrete multi-item recommendations. Within this framework, we couple producer-side exposure guarantees with a consumer-group fairness objective and explicit business-oriented constraints. Our experiments show that the previously reported ``free fairness'' regime from highly stylized single-item recommendation settings disappears once each consumer receives multiple recommendations, and that moderate producer-fairness constraints can improve simulated business metrics by diversifying exposure away from saturated producers. We further show that reduction of inter-group disparity, preserves competitive overall utility.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。