Intersectional Two-sided Fairness in Recommendation

TL;DR

This work targets intersectional two-sided fairness in Top-N recommender systems, showing that unfairness can persist even when both user and item fairness are individually satisfied. It introduces ITFR, a three-component method comprising sharpness-aware disadvantage-group discovery, collaborative loss balance, and predicted score normalization to balance cross-group learning and align positive predictions. Across three public datasets, ITFR substantially reduces intersectional two-sided unfairness while maintaining competitive accuracy, and it remains compatible with fairness-aware reranking methods. The paper advances practical fairness for multi-stakeholder RS by addressing intersectional group effects and providing a plug-in approach adaptable to multiple attributes.

Abstract

Fairness of recommender systems (RS) has attracted increasing attention recently. Based on the involved stakeholders, the fairness of RS can be divided into user fairness, item fairness, and two-sided fairness which considers both user and item fairness simultaneously. However, we argue that the intersectional two-sided unfairness may still exist even if the RS is two-sided fair, which is observed and shown by empirical studies on real-world data in this paper, and has not been well-studied previously. To mitigate this problem, we propose a novel approach called Intersectional Two-sided Fairness Recommendation (ITFR). Our method utilizes a sharpness-aware loss to perceive disadvantaged groups, and then uses collaborative loss balance to develop consistent distinguishing abilities for different intersectional groups. Additionally, predicted score normalization is leveraged to align positive predicted scores to fairly treat positives in different intersectional groups. Extensive experiments and analyses on three public datasets show that our proposed approach effectively alleviates the intersectional two-sided unfairness and consistently outperforms previous state-of-the-art methods.

Figures (8)

• Figure 1: Illustration of intersectional two-sided unfairness. In this toy example, the RS strategy meets two-sided fairness but shows unfair in an intersectional two-sided view. The thumb-up means the recommendation fits the user interest.
• Figure 2: Illustration for our method. $d$ denotes the difference in predicted scores between two interactions.
• Figure 3: Ablation for three components in our method.
• Figure 4: Ablation for two goals in our method.
• Figure 5: Reranking results on the Tenrec dataset. Results on other datasets are similar and omitted. All metrics are the lower the better except for NDCG@20.

Theorems & Definitions (2)

• definition 1: Intersectional Two-sided Group
• definition 2: $\epsilon-$Intersectional Two-sided Fairness