Behavior-Based Address Clustering in Account-Based Blockchains

Level: MA
Responsible person: Syed Muhammad Yasir
Keywords:  Blockchain Analytics, Clustering Heuristics, Account-Based Blockchains, Graph Representation Learning, Entity Identification

Public blockchains such as Ethereum operate on an account-based model, where users can create multiple pseudonymous addresses at no cost. This makes it difficult to estimate the true number of unique entities interacting on-chain, affecting transparency and identity-based applications. Address clustering techniques aim to group blockchain addresses controlled by the same entity, providing insights into user behavior, transaction patterns, and network influence without requiring off-chain identity disclosures. 

This thesis investigates behavior-based clustering techniques for account-based blockchains, specifically targeting transaction graph structures and heuristics that link addresses based on interaction patterns. The research focuses on clustering methods that rely on graph representation learning, network topology, and asset transfer patterns to infer connections between blockchain addresses. 

• Extract and preprocess blockchain data (Ethereum or similar PoS-based networks) for clustering analysis, using transaction records and token transfers. 

• Apply graph-based clustering heuristics such as graph representation learning (Diff2Vec, Role2Vec, DeepWalk) to group similar addresses. 

• Develop and evaluate heuristic-based clustering approaches by analyzing self-authorization patterns, asset transfers, and token interactions, and compare their effectiveness in estimating real-world entity counts using ground truth data (e.g., ENS-linked address pairs). 

• Develop a behavioral model for address interactions, focusing on transaction frequency, counterparties, and economic activity. 

References: 

[1] Wu, Y., & Liu, Z. (2017). Behavior Pattern Clustering in Blockchain Networks. Multimedia Tools and Applications, 76(24), 25743-25760. DOI: 10.1007/s11042-017-4396-4 
[2] Harlev, M., Sun Yin, H., Langenheldt, K., & Mukkamala, R. R. (2019). Clustering Blockchain Data: Techniques, Toolboxes, and Applications. In Machine Learning for Blockchain Applications: Approaches and Challenges (pp. 127-153). Springer.  
[3] Pham, H. Q., Yoneki, E., & Ghiassi-Farrokhfal, Y. (2020). Detecting Malicious Accounts in Permissionless Blockchains using Temporal Graph Properties. arXiv preprint arXiv:2007.05169. 
[4] Thürkauf, D. (2023). Address Clustering Heuristics for Account-Based Blockchain Networks: An Analysis based on a Decentraland User Set. SSRN Electronic Journal. DOI: 10.2139/ssrn.4589925