A Fair Evaluation of Graph Foundation Models for Node Property Prediction

This study conducts a rigorous reevaluation of nine recent Graph Foundation Models (GFMs) for node property prediction to address the lack of unified evaluation standards in the field. The authors compare these models against strong Graph Neural Network (GNN) baselines to determine their relative performance and efficiency.

• The research evaluates nine recent GFMs specifically designed for node property prediction tasks, which are critical for applications like fraud detection and recommendation systems.
• The study compares these GFMs against well-tuned GNN baselines using a fair and rigorous methodology to enable reliable comparison.
• Only the most recent GFMs based on the Prior-data Fitted Networks paradigm were found to outperform the strong GNN baselines in predictive performance.
• While the Prior-data Fitted Network-based models achieve superior accuracy, they do so at a higher inference cost compared to the baselines.

The findings indicate that while specific new architectures can surpass traditional GNNs, practitioners must weigh this performance gain against increased computational costs when selecting models for real-world applications.