NeuBM: Mitigating Model Bias in Graph Neural Networks through Neutral Input Calibration

TL;DR

NeuBM addresses class-imbalance bias in graph neural networks by post-hoc calibration using a neutral graph as a balanced reference. It constructs $G_{\text{neutral}}$ from dataset-wide statistics, computes neutral logits $L_{\text{neutral}}$, and calibrates input logits via $L_{\text{corrected}} = L - L_{\text{neutral}}$, followed by $\\hat{y} = \\text{softmax}(L_{\text{corrected}})$. Empirical results across eight benchmark graphs show consistent gains in F1-macro and minority-class recall with only modest overhead, and ablations confirm the neutral graph and subtraction-based calibration are critical. The approach is architecture-agnostic and scalable, offering practical fairness improvements for real-world imbalanced graph data with minimal retraining.

Abstract

Graph Neural Networks (GNNs) have shown remarkable performance across various domains, yet they often struggle with model bias, particularly in the presence of class imbalance. This bias can lead to suboptimal performance and unfair predictions, especially for underrepresented classes. We introduce NeuBM (Neutral Bias Mitigation), a novel approach to mitigate model bias in GNNs through neutral input calibration. NeuBM leverages a dynamically updated neutral graph to estimate and correct the inherent biases of the model. By subtracting the logits obtained from the neutral graph from those of the input graph, NeuBM effectively recalibrates the model's predictions, reducing bias across different classes. Our method integrates seamlessly into existing GNN architectures and training procedures, requiring minimal computational overhead. Extensive experiments on multiple benchmark datasets demonstrate that NeuBM significantly improves the balanced accuracy and recall of minority classes, while maintaining strong overall performance. The effectiveness of NeuBM is particularly pronounced in scenarios with severe class imbalance and limited labeled data, where traditional methods often struggle. We provide theoretical insights into how NeuBM achieves bias mitigation, relating it to the concept of representation balancing. Our analysis reveals that NeuBM not only adjusts the final predictions but also influences the learning of balanced feature representations throughout the network.

Figures (10)

• Figure 1: Visualization of the impact of class imbalance and Neutral Graph Calibration on GNN predictions, illustrated on the Cora dataset. Left: Original data distribution showing a moderate imbalance across classes. Middle: Biased GNN predictions exhibiting significant misclassifications, especially for minority classes. Right: Predictions after applying NeuBM, demonstrating improved classification accuracy and reduced bias across all classes.
• Figure 2: Class-wise F1-scores on Cora dataset
• Figure 3: Scalability analysis of NeuBM compared to GCN
• Figure 4: Performance of different calibration approaches on Cora dataset
• Figure 5: Sensitivity analysis of scaling factor $\lambda$ on Cora dataset

Theorems & Definitions (8)

• Theorem 1: Bias Reduction
• proof
• Theorem 2: Minority Class Improvement
• proof
• Theorem 3: Stability
• proof
• Theorem 4: Representation Balance
• proof