Bitcoin transaction graphs: training, labelled entities, and CoinJoins

Address-transaction graphs for machine learning on Bitcoin: training graphs with CSV node features, edge features and heuristic clustering labels, plus per-transaction subgraphs with known entity labels and CoinJoin subgraphs as Parquet tables with JSON metadata. Built by the authors from blockchain data and address-clustering heuristics as companion data for a graph neural network clustering method. Useful for training and benchmarking address clustering; heuristic labels may wrongly merge distinct users.

Category

Transactions

Format

CSV, Parquet, JSON

License

CC BY 4.0 (data), not specified (code)

Author

Hugo Schnoering and Michalis Vazirgiannis (LIX, École Polytechnique)

DOI

10.5281/zenodo.22239037

Tags

• addresses
• coinjoin
• machine-learning
• transactions

Links

• btc-graph-hc (code)
• Zenodo record
• Refining Heuristic-Based Bitcoin Address Clustering with Graph Neural Networks (arXiv)