Let's Grow an Unbiased Community: Guiding the Fairness of Graphs via New Links

TL;DR

This work tackles fairness in graph neural networks by guiding the growth of biased real-world graphs through the selective addition of a limited number of links. It introduces FairGuide, a bi-level optimization framework that uses a differentiable, pseudo-downstream task based on community detection and meta-gradient driven link addition to identify edges that reduce structural bias while preserving task performance. The authors provide theoretical justification that mitigating bias in the pseudo-task correlates with reduced bias in downstream predictions, and they validate the approach on large real-world datasets (Github and Pokec) showing consistent improvements in fairness metrics across node-classification and community-detection tasks, with a reasonable trade-off in utility. The method generalizes across GNN backbones and maintains efficiency via a decoupled, differentiable community detection step and an edge-sampling scheme using Gumbel-max; this offers a practical, budget-aware solution for enhancing fairness in graph-based applications.

Abstract

Graph Neural Networks (GNNs) have achieved remarkable success across diverse applications. However, due to the biases in the graph structures, graph neural networks face significant challenges in fairness. Although the original user graph structure is generally biased, it is promising to guide these existing structures toward unbiased ones by introducing new links. The fairness guidance via new links could foster unbiased communities, thereby enhancing fairness in downstream applications. To address this issue, we propose a novel framework named FairGuide. Specifically, to ensure fairness in downstream tasks trained on fairness-guided graphs, we introduce a differentiable community detection task as a pseudo downstream task. Our theoretical analysis further demonstrates that optimizing fairness within this pseudo task effectively enhances structural fairness, promoting fairness generalization across diverse downstream applications. Moreover, FairGuide employs an effective strategy which leverages meta-gradients derived from the fairness-guidance objective to identify new links that significantly enhance structural fairness. Extensive experimental results demonstrate the effectiveness and generalizability of our proposed method across a variety of graph-based fairness tasks.

Figures (8)

• Figure 1: Illustration of guiding the fairness of graphs via new links. (a) An example of a biased social network with structural barriers between user subgroups; (b) Introduction of new links among subgroups to foster unbiased community.
• Figure 2: An illustration of which add new links to guide the growing of graph structures toward fairness.
• Figure 3: Fairness improvements on different GNNs.
• Figure 4: Visualization of trade-off between utility and fairness. Methods in the upper-left region are better.
• Figure 5: Impacts of number of added links.

Theorems & Definitions (11)

• Definition 1: Statistical Parity
• Definition 2: Equal Opportunity
• Definition 3
• Theorem 1
• proof
• Lemma 1
• proof
• Lemma 2
• proof
• Theorem 2