Network Science and Graph Embedding Benchmarks from Benedek Rozemberczki

Graph-structured benchmarks for node classification and link prediction were curated by Benedek Rozemberczki, with the latest update in December 2025. The repository provides data for evaluating algorithms in network science and machine learning research.

Use Cases

• Node classification using graph-convolutional networks (GCN) to predict categorical node labels
• Link prediction to identify missing or future edges within the network structure
• Community detection to identify clusters of nodes based on network-science metrics

Strengths

• Curated specifically for graph neural network (GNN) benchmarking
• Supports multiple embedding techniques including node2vec, deepwalk, and graph2vec
• Released under the permissive MIT license

Limitations

• Lack of a unified schema across all included datasets
• Specific feature descriptions for every graph are not provided in the top-level metadata

Provenance

Source

Benedek Rozemberczki via GitHub

Collection Method

Curated from various network science research sources

Freshness

Last updated December 2025.