Graph Learning-Based Blockchain Phishing Account Detection with a Heterogeneous Transaction Graph

Abstract

Recently, cybercrimes that exploit the anonymity of blockchain are increasing. They steal blockchain users’ assets, threaten the network’s reliability, and destabilize the blockchain network. Therefore, it is necessary to detect blockchain cybercriminal accounts to protect users’ assets and sustain the blockchain ecosystem. Many studies have been conducted to detect cybercriminal accounts in the blockchain network. They represented blockchain transaction records as homogeneous transaction graphs that have a multi-edge. They also adopted graph learning algorithms to analyze transaction graphs. However, most graph learning algorithms are not efficient in multi-edge graphs, and homogeneous graphs ignore the heterogeneity of the blockchain network. In this paper, we propose a novel heterogeneous graph structure called an account-transaction graph, ATGraph. ATGraph represents a multi-edge as single edges by considering transactions as nodes. It allows graph learning more efficiently by eliminating multi-edges. Moreover, we compare the performance of ATGraph with homogeneous transaction graphs in various graph learning algorithms. The experimental results demonstrate that the detection performance using ATGraph as input outperforms that using homogeneous graphs as the input by up to 0.2 AUROC.

Dynamic Graph Representation Learning With Neural Networks: A Survey

“…The second combination alternates GNN layers and Transformer layers in the DG encoder [100], [126], [136]. The third combination is a parallel encoding of the DG by independent transformer and GNN layers, followed by a combination of their encoded hidden states [120], merging the strengths of both layers. Additionally, some dynamic graph models [73], [98], [99], [122], [123], [125], [127] use transformers exclusively as graph encoders, exploiting the self-attention mechanism for node hidden states propagation without relying on traditional GNN architectures.…”

“…We have distinguished position encoding into 4 types. The first method is the embedding based on the time-related index of nodes or edges [98], [100], [120], [124], [127]. The second method utilizes (global) graph spectral information [99], such as the Laplacian or its eigenvectors.…”

“…Concerning Self-Attention (SA), there are four solutions on dynamic graphs, as shown in Fig. 12 (right): Spatial SA: when a token represents a node in a snapshot, SA operates at the spatial level, equivalent to the graph encoder f G , with SA matrix dimensions being the square of the number of nodes N t in the snapshot t [101], [120]- [123], [125], [136].…”

“…Temporal SA: When a token represents the same node (or the entire snapshot) across multiple snapshots, SA operates at the temporal level and acts as a temporal encoder f T , its SA matrix dimensions thus are the square of the number of time steps K [101], [120], [123], [125], [136]. Spatial-Temporal SA: Following Temporal Edge Modeling III-B, creating replicas for each node in K snapshots, forming Spatial-Temporal Attention, with the Attention matrix size being the square of N t or K × T [98]- [101], [139].…”

Development of Dag Blockchain Model

“…These algorithms fall broadly into three distinct categories, each with its unique approach to achieving consensus within the network. The first category encompasses proof-based consensus algorithms, wherein nodes seeking to participate in the verification process must demonstrate their qualification for appending tasks [9]. The second category revolves around voting-based consensus mechanisms, requiring validators to share their validation results before reaching a final decision on new blocks 33 DOI: 10.37943/16CGOY7609 © Akylbek Tokhmetov, Vyacheslav Lee, Liliya Tanchenko or transactions.…”