Detecting Malicious Blockchain Transactions Using Graph Neural Networks

Detecting Malicious Blockchain Transactions Using Graph Neural Networks

연구 분야: Strategies

학회: International Symposium on Distributed Ledger Technology

The adoption of blockchain technology within various critical infrastructures is on the rise. Concurrently, there has been a corresponding increase in its misuse, primarily through the exploitation of its pseudo-anonymous characteristic. Encouraging blockchain adoption and improving security in the decentralised environment require techniques to detect wallets and/or smart contracts owned by malicious entities. Illegal activities such as dark market trades, money laundering, and receiving unlawful payments are performed by connecting various wallets or smart contracts in a meticulous way. A graph can be a potential representation to visualise such interconnections via various patterns, and graph-based data may represent the topological structure of the blockchain network. Recently, Graph Neural Networks (GNN) have been widely used for analysing the structure of complex networks and identifying patterns. This is the first work that considers a generalised graph representation for the Bitcoin and Ethereum networks and analyses their behaviour using a combination of heterogeneous GNN framework’s GraphSAGE and Graph Attention Network (GAT). The classification results reveal that the proposed approach modestly improved Bitcoin network analysis, whereas Ethereum smart contract analysis needs further investigation in terms of incorporating other aspects of smart contracts, such as code-base, byte length, and lifetime features.