Regret-aware Re-ranking for Guaranteeing Two-sided Fairness and Accuracy in Recommender Systems

TL;DR

The paper tackles fairness in two-sided recommender systems by addressing not only average user accuracy and provider exposure but also individual user fairness, a gap in prior work. BankFair+ combines a regret theory-based, non-linear user satisfaction model with regret-aware fuzzy programming and a bankruptcy-problem–driven exposure allocation to balance competing goals online. It introduces Module1 for time-sequenced minimum exposure allocation via the Talmud rule and Module2 for regret-aware online re-ranking, including an online dual learning mechanism. Empirical results on KuaiRand-1K and Huawei-Video demonstrate Pareto-optimal improvements in average accuracy, provider fairness, and notably reduced disparities across users, validating the approach's practical impact for dynamic, multi-stakeholder RS.

Abstract

In multi-stakeholder recommender systems (RS), users and providers operate as two crucial and interdependent roles, whose interests must be well-balanced. Prior research, including our work BankFair, has demonstrated the importance of guaranteeing both provider fairness and user accuracy to meet their interests. However, when they balance the two objectives, another critical factor emerges in RS: individual fairness, which manifests as a significant disparity in individual recommendation accuracy, with some users receiving high accuracy while others are left with notably low accuracy. This oversight severely harms the interests of users and exacerbates social polarization. How to guarantee individual fairness while ensuring user accuracy and provider fairness remains an unsolved problem. To bridge this gap, in this paper, we propose our method BankFair+. Specifically, BankFair+ extends BankFair with two steps: (1) introducing a non-linear function from regret theory to ensure individual fairness while enhancing user accuracy; (2) formulating the re-ranking process as a regret-aware fuzzy programming problem to meet the interests of both individual user and provider, therefore balancing the trade-off between individual fairness and provider fairness. Experiments on two real-world recommendation datasets demonstrate that BankFair+ outperforms all baselines regarding individual fairness, user accuracy, and provider fairness.

Figures (9)

• Figure 1: Although (a) previous works can guarantee both the user average accuracy and provider fairness, (b) severe individual unfairness of user accuracy still exists. (c) With the increase in provider fairness degree, such individual-level unfairness tends to escalate. (d) The remaining issues of existing methods: how to ensure individual fairness while guaranteeing provider fairness and user accuracy.
• Figure 2: (1) Workflow of the proposed BankFair+ re-ranking algorithm. (b) In module 1, we allocate exposure to providers using the Talmud rule in bankruptcy problem. (c) In module 2, we propose a regret-aware online re-ranking algorithm to mitigate individual unfairness.
• Figure 3: The Pareto frontierof user accuracy (NDCG@K) and provider fairness (Gini@K and ESP@K) on two different datasets with top-$10$. X-axis shows Gini@K metric and ESP@K metric, while Y-axis shows NDCG@K metric. $\uparrow$ means higher values are better and $\downarrow$ favors lower values.
• Figure 4: The Pareto frontier of individual accuracy (MMR@K) and provider fairness (Gini@K and ESP@K) on two different datasets with top-$10$. X-axis shows Gini@K metric and ESP@K metric, while Y-axis shows MMR@K metric. $\uparrow$ means higher values are better and $\downarrow$ favors lower values.
• Figure 5: The user distribution on individual accuracy of our method BankFair+ and the best-performed baselines. The hyperparameters of the method were tuned to achieve the best average user accuracy with 75% ESP@10. The red line indicates the average accuracy level of users.

Theorems & Definitions (2)

• Theorem 1
• Theorem 2