Fairness for Robust Learning to Rank

TL;DR

This paper introduces a fairness-aware ranking system based on distributional robustness, formulating a minimax game to balance utility and fairness, resulting in improved utility for fair rankings.

Contribution

It presents a novel distributionally robust approach to fairness in learning to rank, combining minimax optimization with fairness constraints.

Findings

Better utility for highly fair rankings compared to baselines

Effective balancing of utility and fairness in ranking systems

Robustness against adversarial distribution shifts

Abstract

While conventional ranking systems focus solely on maximizing the utility of the ranked items to users, fairness-aware ranking systems additionally try to balance the exposure for different protected attributes such as gender or race. To achieve this type of group fairness for ranking, we derive a new ranking system based on the first principles of distributional robustness. We formulate a minimax game between a player choosing a distribution over rankings to maximize utility while satisfying fairness constraints against an adversary seeking to minimize utility while matching statistics of the training data. We show that our approach provides better utility for highly fair rankings than existing baseline methods.