Dynamic fairness-aware recommendation through multi-agent social choice

TL;DR

This work tackles the challenge of dynamic, multi-stakeholder fairness in personalized recommendations by introducing SCRUF-D, a two-stage social-choice framework that models fairness concerns as separate agents. The architecture separates allocation (which fairness agents participate) from the choice phase (how agent preferences and user preferences are merged), enabling multiple, potentially conflicting fairness objectives to be addressed in real time. Extensive experiments on MovieLens, Microlending, and synthetic data show how different allocation and choice mechanisms trade off accuracy and fairness, with dynamic, context-aware decisions often outperforming static baselines. The framework offers a flexible path toward integrating diverse fairness notions into recommender systems without re-training underlying models, with practical implications for platforms like Kiva and beyond.

Abstract

Algorithmic fairness in the context of personalized recommendation presents significantly different challenges to those commonly encountered in classification tasks. Researchers studying classification have generally considered fairness to be a matter of achieving equality of outcomes between a protected and unprotected group, and built algorithmic interventions on this basis. We argue that fairness in real-world application settings in general, and especially in the context of personalized recommendation, is much more complex and multi-faceted, requiring a more general approach. We propose a model to formalize multistakeholder fairness in recommender systems as a two stage social choice problem. In particular, we express recommendation fairness as a novel combination of an allocation and an aggregation problem, which integrate both fairness concerns and personalized recommendation provisions, and derive new recommendation techniques based on this formulation. Simulations demonstrate the ability of the framework to integrate multiple fairness concerns in a dynamic way.

Figures (8)

• Figure 1: SCRUF-D Framework / Allocation Phase: Recommendation opportunities are allocated to fairness concerns based on the context.
• Figure 2: SCRUF-D Framework / Choice Phase: The preferences derived from the recommender system and the fairness concerns are integrated by the choice mechanism.
• Figure 3: Accuracy vs fairness for the MovieLens dataset at different values of $\lambda$.
• Figure 4: Comparison of mechanisms on MovieLens data. OFair is omitted as it is far off the chart to the lower left. Multi-FR is far below in terms of accuracy.
• Figure 5: Accuracy vs fairness for the Microlending dataset at different values of $\lambda$. Value for Weighted + Borda had much lower fairness and were omitted.