Equity by Design: Fairness-Driven Recommendation in Heterogeneous Two-Sided Markets

TL;DR

This work addresses fairness in heterogeneous two-sided marketplaces by transitioning from soft, single-item allocations to discrete multi-item recommendations and by introducing group fairness via Conditional Value at Risk (CVaR). It integrates business constraints through a Gross Merchandise Value (GMV) threshold and develops scalable solvers (LP relaxation with rounding and gradient-based methods) to enable production-scale fairness. Empirically, the classical 'free fairness' regime vanishes for multi-item settings, with fairness imposing a measurable cost on consumer utility at larger k, while CVaR reduces disparities across consumer groups and can improve business metrics when paired with moderate exposure constraints. The framework provides a practical, scalable pathway to align consumer relevance, producer exposure, and revenue in real-world platforms, and highlights the importance of group-level fairness in heterogeneous populations.

Abstract

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 real platforms face finer-grained complexity: consumers differ in preferences and engagement patterns, producers vary in catalog value and capacity, and business objectives impose additional constraints beyond raw relevance. We formalize two-sided fairness under these realistic conditions, extending prior work from soft single-item allocations to discrete multi-item recommendations. We introduce Conditional Value-at-Risk (CVaR) as a consumer-side objective that compresses group-level utility disparities, and integrate business constraints directly into the optimization. Our experiments reveal that the "free fairness" regime, where producer constraints impose no consumer cost, disappears in multi item settings. Strikingly, moderate fairness constraints can improve business metrics by diversifying exposure away from saturated producers. Scalable solvers match exact solutions at a fraction of the runtime, making fairness-aware allocation practical at scale. These findings reframe fairness not as a tax on platform efficiency but as a lever for sustainable marketplace health.

Figures (9)

• Figure 1: Fairness-aware re-ranking for heterogeneous two-sided markets. An upstream model produces consumer--producer relevance scores $\rho$. Our method computes discrete multi-item allocations that jointly optimize consumer group fairness (CVaR), producer exposure guarantees, and business constraints.
• Figure 2: Runtime comparison of various optimization methods on the Amazon Reviews dataset under $\gamma = 0.5$, $\alpha = 0.95$, and $k = 10$ as a function of increasing relevance matrix size.
• Figure 3: Mean consumer utility versus producer fairness ($\gamma$) for $k \in \{1, 5, 10\}$ across Amazon Reviews, MovieLens, and SimRec. At $k=1$, utility remains nearly flat as $\gamma$ increases (free fairness). At $k>1$, utility declines with stricter fairness, and the slope steepens with larger $k$.
• Figure 4: Group-level consumer utilities under max-min, mean, and CVaR optimization across $\gamma$. Lines represent consumer groups, colored by share of total sample (darker = larger).
• Figure 5: Simulated sell-through rate (STR) as a function of producer fairness level $\gamma$ using mean and CVaR consumer utility objectives.