Towards Controllable Hybrid Fairness in Graph Neural Networks

摘要

Graph Neural Networks (GNNs) have shown remarkable capabilities in mining graph-structured data. However, conventional GNNs often encounter various fairness issues, such as predictions with prejudices when dealing with nodes with different sensitive attributes like genders or races, or significantly different prediction performance when facing nodes with different degrees. Existing studies mainly focus on addressing one specific fairness issue, neglecting the fact that a GNN model may face multiple unfairness simultaneously in reality, and addressing only one specific fairness may still leave the GNNs in an unfair status. In this paper, we focus on achieving multiple fairness on GNNs simultaneously, which we call hybrid fairness. To achieve this objective, we propose a novel GNN framework called LibraGNN. Specifically, we adopt a multi-teacher knowledge distillation training framework that successfully unifies the learning paradigms for multiple fairness. To ensure LibraGNN strikes a better trade-off among different fairness, we transform the multi-teacher knowledge distillation into a multi-objective optimization problem and further employ Pareto efficiency for optimization guidance. Finally, a controllable preference vector is introduced to assist LibraGNN in modulating its capability towards various forms of fairness, thereby achieving controllable hybrid fairness. Extensive experiments on three real-world datasets demonstrate the effectiveness of LibraGNN on both hybrid fairness and utility.