Degree Bias

Degree bias in graph neural networks (GNNs) refers to the disproportionate performance of GNNs on nodes with varying degrees (number of connections), often favoring high-degree nodes. Current research focuses on mitigating this bias through various methods, including model-agnostic approaches that adjust node representations, community-aware graph transformers that leverage network structure, and test-time augmentations that enhance predictions for low-degree nodes. Addressing degree bias is crucial for ensuring fairness and improving the reliability of GNNs across diverse applications, such as recommendation systems and social network analysis, where equitable treatment of all nodes is paramount.