Paper Overview
Field: Machine Learning Authors: Valentijn Oldenburg, Floris de Kam, Stef de Wildt Published: 2026-08-12 arXiv: 2508.05138
Summary
In fair ranked link prediction, demographic parity (\(\Delta_\mathrm{DP}\)) 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, the authors reproduce this claim by showing that \(\Delta_\mathrm{DP}\) 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.
They also reproduce the effectiveness of MORAL, a post-processing method that improves exposure-based fairness while maintaining competitive utility. Beyond reproduction, they assess robustness using synthetic homophily settings, categorical sensitive attributes, and additional fairness and utility metrics, including pair-adaptive attention-weighted ranking fairness (AWRF).
Key Findings
- \(\Delta_\mathrm{DP}\) can hide bias: it may show parity overall while certain subgroup-pair links are systematically ranked lower.
- NDKL, a rank-aware metric, successfully detects these exposure disparities.
- MORAL reduces exposure-based bias with minimal utility loss across diverse settings and datasets.
- Findings hold under synthetic homophily settings, categorical sensitive attributes, and additional metrics including AWRF.
- Paper: https://arxiv.org/abs/2508.05138
- Corrected reproducible implementation: https://github.com/Floris93100/reproducing-MORAL