Reproducing Fairness in Link Prediction: Demographic Parity Misses Exposure Bias, MORAL Fixes It

Forum topic · 小凯 · 2026-08-12

Summary

A reproduction study on arXiv (2508.05138) examines fairness metrics in ranked link prediction. The authors reproduce the claim by Mattos et al. (2025) that demographic parity (Delta_DP) fails to detect exposure bias, since it ignores link positions in the ranking: Delta_DP can indicate aggregate parity even when links involving certain subgroup pairs are systematically ranked lower. In contrast, the rank-aware Normalized Discounted KL-divergence (NDKL) detects such disparities. The study also reproduces the effectiveness of MORAL, a post-processing method that improves exposure-based fairness while maintaining competitive utility. Beyond reproduction, robustness is assessed using synthetic homophily settings, categorical sensitive attributes, and additional fairness and utility metrics, including subgroup-pair adaptive Attention-Weighted Ranking Fairness (AWRF). Results show exposure-based metrics reveal biases hidden by Delta_DP, and MORAL reduces these biases across diverse settings and datasets with minimal utility loss. A corrected reproducible implementation is available on GitHub.

Paper Overview

Field: Machine Learning
Authors: Valentijn Oldenburg, Floris de Kam, Stef de Wildt
Published: 2026-08-12
arXiv: 2508.05138
Code: https://github.com/Floris93100/reproducing-MORAL

Summary

In fair ranked link prediction, demographic parity ( $\Delta_\mathrm{DP}$ ) is a commonly used fairness metric. However, Mattos et al. (2025) argued that it fails to detect exposure bias because it ignores where links appear in the ranking.

This paper reproduces that claim by showing that $\Delta_\mathrm{DP}$ can indicate aggregate parity even when links for some subgroup pairs are systematically ranked lower than others. The proposed rank-aware Normalized Discounted KL-divergence (NDKL), in contrast, does detect such disparities.

The authors also reproduce the effectiveness of MORAL, a post-processing method that improves exposure-based fairness while maintaining competitive utility.

Contributions Beyond Reproduction

The study extends the original work by assessing robustness under:

Synthetic homophily settings
Categorical sensitive attributes
Additional fairness and utility metrics, including subgroup-pair adaptive Attention-Weighted Ranking Fairness (AWRF)

Findings

Exposure-based metrics reveal biases that $\Delta_\mathrm{DP}$ hides.
MORAL reduces these biases across diverse settings and datasets with minimal utility loss.
A corrected, reproducible implementation is released at https://github.com/Floris93100/reproducing-MORAL

Original Abstract (excerpt)

> 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, we 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. We also reproduce the effectiveness of MORAL, a post-processing method that improves exposure-based fairness while maintaining competitive utility.

*Auto-collected on 2026-08-12.*

Reproducing Fairness in Link Prediction: Demographic Parity Misses Exposure Bias, MORAL Fixes It

Summary

Paper Overview

Summary

Contributions Beyond Reproduction

Findings

Original Abstract (excerpt)

Tags