Fairness in Link Prediction Beyond Demographic Parity: A Reproducibility Study
2026-08-10 • Machine Learning
Machine LearningSocial and Information Networks
AI summaryⓘ
The authors reproduced previous work showing that a common fairness measure in ranked link prediction, demographic parity (Δ_DP), can miss important biases because it doesn't consider the position of links in the ranking. They confirmed that a new rank-aware metric, Normalized Discounted KL-divergence (NDKL), can detect these position-based biases better. They also verified that MORAL, a fairness-improving method, effectively reduces bias related to exposure without significantly hurting overall performance. Additionally, they tested these findings across various data setups and fairness metrics to confirm robustness and shared their reproducible code online.
fair ranked link predictiondemographic parityexposure biasNormalized Discounted KL-divergence (NDKL)MORALpost-processing fairnesshomophilyAttention-Weighted Rank Fairness (AWRF)utilityfairness metrics
Authors
Valentijn Oldenburg, Floris de Kam, Stef de Wildt, Jarno Nilson Balk
Abstract
In fair ranked link prediction, demographic parity ($Δ_\mathrm{DP}$) is a common fairness metric. Yet, Mattos et al. (2025) argue that it fails to detect exposure bias because it ignores where links appear in the ranking. In this study, we reproduce this claim by showing that $Δ_\mathrm{DP}$ can indicate aggregate parity even when some subgroup-pair links are systematically ranked lower than others. The proposed rank-aware Normalized Discounted KL-divergence (NDKL), however, does detect such disparities. We also reproduce the effectiveness of MORAL, a post-processing method that improves exposure-based fairness while maintaining competitive utility. Beyond reproduction, we assess robustness using synthetic homophily settings, categorical sensitive attributes, and additional fairness and utility metrics, including subgroup-pair-adapted Attention-Weighted Rank Fairness (AWRF). Overall, our results show that exposure-based metrics uncover biases hidden by $Δ_\mathrm{DP}$ and that MORAL reduces these biases with minimal utility loss across diverse settings and datasets. We release a corrected, reproducible implementation at https://github.com/Floris93100/reproducing-MORAL.