Abstract

Bitcoin is a digital currency where transactions between users are recorded on a public ledger, known as the blockchain. We explore a subset of transactional data from the Bitcoin blockchain during the first four years of its existence. Our goal is to identify communities of related users and their behavioural spending-patterns. To that end, we represent this dataset as a temporal network of users, with weighted edges signifying the transfer of bitcoin amongst users at a certain time. We construct a Bayesian nonparametric mixture model for discovering latent class-structure in transactional data networks. Furthermore, we approximate the posterior distribution of user partitions using a Metropolis-Hastings algorithm.