Towards Model-Agnostic Post-Hoc Adjustment for Balancing Ranking Fairness and Algorithm Utility

Algorithmic fairness has received lots of interests in machine learning recently. In this paper, we focus on the bipartite ranking scenario, where the instances come from either the positive or negative class and the goal is to learn a ranking function that ranks positive instances higher than negative ones. In an unfair setting, the probabilities of ranking the positives higher than negatives are different across different protected groups. We propose a general post-processing framework, xOrder, for achieving fairness in bipartite ranking while maintaining the algorithm classification performance. In particular, we optimize a weighted sum of the utility and fairness by directly adjusting the relative ordering across groups. We formulate this problem as identifying an optimal warping path across different protected groups and solve it through a dynamic programming process. xOrder is compatible with various classification models and applicable to a variety of ranking fairness metrics. We evaluate our proposed algorithm on four benchmark data sets and one real world patient electronic health record repository. The experimental results show that our approach can achieve great balance between the algorithm utility and ranking fairness. Our algorithm can also achieve robust performance when training and testing ranking score distributions are significantly different.