Research map / Graph learning / Knowledge and temporal graphs
The Temporal Graph of Bitcoin Transactions
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Builds a temporal heterogeneous graph of the complete Bitcoin transaction history up to block 863,000 (about 2.4B nodes and 39.72B edges), with sampling tools, Neo4j snapshots, and statistical profiling of on-chain metrics.
Problem
Bitcoin's UTxO design and pseudonymity make its ledger hard to use for machine learning, since wallet-level state and entity relationships must be reconstructed from raw transaction data.
Approach
Transform the TxOut-centric ledger into a wallet-centric heterogeneous temporal graph with Script, Tx, Block and Coinbase nodes, script-to-script Transfers and Fee edges, and block-height timestamps, then release it with sampling algorithms and graph database snapshots.
Key results
- The graph contains >2.4B nodes and >39.72B edges across 4 node types and 6 edge types, covering blocks up to height 863,000.
- Blockchain profiling shows blocks every ~10 minutes (mean 9.59 ± 13.45 min) and 10.4% (89,714) empty blocks.
- Compressed TSV files exceed 1.17 TB and expand to over 3 TB in a graph database; a database snapshot is over 700 GB.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| represent transaction input and output scripts by their derived addresses instead of separate unique input/output IDs | Bitcoin transaction graph · Bitcoin graph transactions | flow-of-funds representation | Helped |
Surprising findings
- About 10.4% of blocks are empty, containing only the coinbase transaction.
- Script nodes have a much longer degree tail than Tx nodes (max in-degree ≈282M for scripts vs ≈8K for Tx), indicating a small set of frequently reused scripts.
Limitations
- Graph is limited to on-chain information; no off-chain entity labels or annotations are included.
- Initial release excludes cryptographic proofs, zero-value transactions, and transactions with both >20 inputs and >20 outputs.
- Frequent updates are needed as the ledger grows daily; incremental update methods are provided but scale is a challenge.
Experimental setup
Datasets: Bitcoin blockchain ledger up to block 863000
Metrics: node count edge count degree distribution storage size
Builds on: Forest Fire model Bitcoin Core
Closest papers by meaning
- Multi-Chain Graphs of Graphs: A New Approach to Analyzing Blockchain Datasets (NeurIPS 2024 · similarity 0.89)
- EX-Graph: A Pioneering Dataset Bridging Ethereum and X (ICLR 2024 · similarity 0.86)
- BlockScan: Detecting Anomalies in Blockchain Transactions (NeurIPS 2025 · similarity 0.83)
- Temporal-Aware Evaluation and Learning for Temporal Graph Neural Networks (AAAI 2025 · similarity 0.82)
- Incremental Topological Ordering and Cycle Detection with Predictions (ICML 2024 · similarity 0.81)
- Graph Anomaly Detection as Dynamical Transport: Training-Free Scoring via Empirical Bayes (NeurIPS 2026 · similarity 0.81)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.