MAPPING: Debiasing Graph Neural Networks for Fair Node Classification with Limited Sensitive Information Leakage

2024-01-23

World wide web (Bussum)

AI Summary

Graph Neural Networks (GNNs) for node classification suffer from bias induced by spurious correlations between graph structure and sensitive attributes, while also posing privacy risks—hindering fair deployment in critical domains such as healthcare and finance. To address this, we propose MAPPING, the first framework that models sensitive information perturbation as a differentiable mapping, enabling joint optimization of fairness and privacy. Our method comprises: (i) gradient-masking-driven adversarial disentanglement of sensitive features; (ii) a differentially private graph message-passing mechanism; and (iii) a unified fairness–privacy optimization objective. Evaluated on multiple real-world graph datasets, MAPPING reduces ΔEO and ΔDP by over 40%, incurs <1.5% accuracy loss, achieves <8% sensitive attribute reconstruction success rate, and provides theoretically guaranteed upper bounds on information leakage.

Technology Category

• Machine Learning: Privacy
• Natural Language Processing: Ethics — Bias, Fairness, Transparency & Privacy
• Computer Vision: Bias, Fairness & Privacy

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Security and Privacy: Data transparency and provenance
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Problem

• Bias Reduction
• Privacy Preservation
• Fair Node Classification

Innovation

• MAPPING
• Bias Reduction
• Privacy Protection