Title: Developing Address Clustering Heuristics for Account-Based Blockchain Networks: An Analysis based on a Specific Address Set
Author: Dario Thürkauf
Supervisor: Prof. Dr. Fabian Schär
Submission: August 28, 2023
python: Contains all the source code used for data collection, analysis, and visualization.figures: Contains all the figures and plots generated during the analysis.data: Contains the small data sets used for the analysis.
Given the negligible cost of creating blockchain addresses, the notion that one address equates to one user is flawed. Therefore, Blockchain Address Clustering emerges as a key tool for estimating the actual number of users. This paper examines heuristics to cluster addresses using Ethereum and Polygon PoS data, analyzing over 470,000 addresses collected from Decentraland over the course of nine months. We assess various existing heuristics, adapt them to our specific research context, and also introduce a new heuristic. Our findings suggest that graph representation learning, particularly the application of an existing node embedding algorithm to an asset transfer network graph, is the most effective approach for clustering a predefined address set. However, the method does not come without its drawbacks, which is why we propose possible extensions for improving clustering performance.