Equity vs. Equality: Optimizing Ranking Fairness for Tailored Provider Needs

Abstract

Ranking plays a central role in connecting users and providers in Information Retrieval (IR) systems, making provider-side fairness an important challenge. While recent research has begun to address fairness in ranking, most existing approaches adopt an equality-based perspective, aiming to ensure that providers with similar content receive similar exposure. However, it overlooks the diverse needs of real-world providers, whose utility from ranking may depend not only on exposure but also on outcomes like sales or engagement. Consequently, exposure-based fairness may not accurately capture the true utility perceived by different providers with varying priorities. To this end, we introduce an equity-oriented fairness framework that explicitly models each provider's preferences over key outcomes such as exposure and sales, thus evaluating whether a ranking algorithm can fulfill these individualized goals while maintaining overall fairness across providers. Based on this framework, we develop EquityRank, a gradient-based algorithm that jointly optimizes user-side effectiveness and provider-side equity. Extensive offline and online simulations demonstrate that EquityRank offers improved trade-offs between effectiveness and fairness and adapts to heterogeneous provider needs.

Figures (6)

• Figure 1: A schematic diagram of horizontal and vertical allocation yang2023vertical with $2$ users and rank list size $K = 3$.
• Figure 2: A comparison of the fairness ranking results between EquityRank_v and PoorK. We first calculate the ratio of the two types of gain weights for each provider in each dataset, $\frac{v_b}{v_e}$. Then, we compute the ratio of the two types of gains assigned to each provider by the two models, $\frac{Gain_b}{Gain_e}$. Each point represents a provider's own $\frac{v_b}{v_e}$ (the x-axis) and its allocated $\frac{Gain_b}{Gain_e}$ under a specific algorithm (the y-axis).
• Figure 3: The unfair.-aNDCG balance curve for each model across different datasets under the offline setting. For models with a hyperparameter $\alpha$, it indicates the highest NDCG an algorithm can achieve when the unfairness value does not exceed a certain threshold. The upper and lefter ($\nwarrow$) a point, the better its corresponding model's performance.
• Figure 4: The unfair.-aNDCG curves in the two special scenarios under the offline setting. "Exp1st" and "Sale1st" represent providers preferring exposure and sales, respectively.
• Figure 5: The unfair.-cNDCG balance curve for each model across different datasets under the online setting.