New study reveals hidden biases in link prediction fairness metrics

This study reproduces and extends research on fairness in link prediction, focusing on exposure bias beyond demographic parity. The authors demonstrate that a rank-aware metric, NDKL, can detect biases missed by demographic parity, and that the MORAL post-processing method effectively reduces these biases with minimal utility loss. The research also assesses the robustness of these findings across various synthetic settings and fairness metrics, including AWRF, and provides a corrected, reproducible implementation. AI

IMPACT Highlights limitations in common fairness metrics for AI, suggesting improved methods for detecting and mitigating bias in link prediction systems.

RANK_REASON The cluster contains an academic paper published on arXiv detailing a reproducibility study of fairness metrics in link prediction. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• arXiv
• AWRF
• Mattos et al.
• NDKL
• Valentijn Oldenburg

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →