TopoFair: Linking Topological Bias to Fairness in Link Prediction Benchmarks

Graph link prediction (LP) plays a critical role in socially impactful applications, such as job recommendation and friendship formation. Ensuring fairness in this task is thus essential. While many fairness-aware methods manipulate graph structures to mitigate prediction disparities, the topological biases inherent to social graph structures remain poorly understood and are often reduced to homophily alone. This undermines the generalization potential of fairness interventions and limits their applicability across diverse network topologies. In this work, we propose a novel benchmarking framework for fair LP, centered on the structural biases of the underlying graphs. We begin by reviewing and formalizing a broad taxonomy of topological bias measures relevant to fairness in graphs. In parallel, we introduce a flexible graph generation method that simultaneously ensures fidelity to real-world graph patterns and enables controlled variation across a wide spectrum of structural biases. We apply this framework to evaluate both classical and fairness-aware LP models across multiple use cases. Our results provide a fine-grained empirical analysis of the interactions between predictive fairness and structural biases. This new perspective reveals the sensitivity of fairness interventions to beyond-homophily biases and underscores the need for structurally grounded fairness evaluations in graph learning.

💡 Research Summary

The paper “TopoFair: Linking Topological Bias to Fairness in Link Prediction Benchmarks” addresses a critical gap in the fairness literature for graph link prediction (LP). While many recent works focus on mitigating unfair outcomes by modifying graph structures or re‑weighting edges, they largely treat the underlying topological bias as a single dimension—homophily. The authors argue that real‑world social graphs exhibit a richer set of structural biases (e.g., degree centrality disparities, neighborhood diversity, information flow asymmetries) that can also drive unfair predictions. To study these effects systematically, the paper makes three major contributions.

First, it proposes a unified taxonomy of structural bias measures. The taxonomy splits biases into node‑level and graph‑level, and further into topological versus flow‑based categories. Node‑level topological metrics include closeness, betweenness, prestige, degree, constraint, density, heterogeneity, and heterophily. Flow‑based node metrics such as effective resistance and information control capture how easily information spreads from a node. Graph‑level topological metrics cover assortativity (homophily), average mixed distance between groups, power‑law exponent ratios (hub concentration), while flow‑based graph metrics quantify information unfairness across groups. For each metric M, the authors define a normalized disparity ωM = (E