FairACE: Achieving Degree Fairness in Graph Neural Networks via Contrastive and Adversarial Group-Balanced Training

TL;DR

This paper tackles degree-based fairness biases in Graph Neural Networks by proposing FairACE, a framework that fuses asymmetric contrastive learning with adversarial training and a group-balanced fairness loss to balance performance across node degree groups. It introduces the Accuracy Distribution Gap (ADG) as a Wasserstein-distance-based metric to quantify distributional fairness across degree-based groups and defines Overall ADG (OADG) for overall assessment. The method combines a graph encoder with online/target EMA-based asymmetric contrastive learning, a GRL-enabled degree discriminator, and a group-balanced loss to ensure degree-agnostic representations while preserving accuracy. Empirical results on four public graphs show substantial improvements in fairness metrics (DSP, DEO, ADG) with competitive classification accuracy, and ablations confirm the necessity of each component for achieving balanced performance in GNNs.

Abstract

Fairness has been a significant challenge in graph neural networks (GNNs) since degree biases often result in un-equal prediction performance among nodes with varying degrees. Existing GNN models focus on prediction accuracy, frequently overlooking fairness across different degree groups. To addressthis issue, we propose a novel GNN framework, namely Fairness- Aware Asymmetric Contrastive Ensemble (FairACE), which inte-grates asymmetric contrastive learning with adversarial training to improve degree fairness. FairACE captures one-hop local neighborhood information and two-hop monophily similarity to create fairer node representations and employs a degree fairness regulator to balance performance between high-degree and low-degree nodes. During model training, a novel group-balanced fairness loss is proposed to minimize classification disparities across degree groups. In addition, we also propose a novel fairness metric, the Accuracy Distribution Gap (ADG), which can quantitatively assess and ensure equitable performance across different degree-based node groups. Experimental results on both synthetic and real-world datasets demonstrate that FairACE significantly improves degree fairness metrics while maintaining competitive accuracy in comparison to the state-of-the-art GNN models.

Figures (3)

• Figure 1: Overview Structure of the Proposed FairACE framework.
• Figure 2: Ablation study on the impact of the main module on classification accuracy.
• Figure 3: Ablation study on the effect of the main module on the ADG metric.