GNN Heterophily: Curated Benchmarks and Resources for Graph Neural Networks

This repository provides a curated collection of graph datasets and research papers focusing on heterophily in Graph Neural Networks, maintained by alexfanjn and updated through December 2024. It serves as a central directory for benchmarking GNN models on graphs where connected nodes frequently belong to different classes.

Use Cases

• Benchmarking GNN architectures against heterophilic graph datasets to test non-assortative learning
• Comparing model performance using homophily ratio metrics provided in the literature
• Literature review of GNN research focusing on node classification in non-homophilous graphs

Strengths

• Updated as of December 2024
• Categorized specifically by heterophily and homophily metrics
• GPL-3.0 licensed repository structure

Limitations

• Does not provide a unified schema as it aggregates various independent datasets
• Metadata for individual datasets like row counts or feature types must be sourced from external links

Provenance

Source

alexfanjn via GitHub

Collection Method

curated collection of research resources and dataset links

Freshness

Last updated December 2024.

Geography

global