RPHunter: Unveiling Rug Pull Schemes in Crypto Token via Code-and-Transaction Fusion Analysis

Description:This is a repository that contains the source code, dataset, and experiment results of our paper in the support of the Research Paper submission.RPHunter is a novel technique that models code risk information and suspicious transaction behavior within two graph structures, utilizing graph neural networks (GNNs) for feature learning to enhance Rug Pull detection across varied scenarios.We have conducted an empirical study to uncover the underlying mechanisms of Rug Pull by analyzing 645 real-world Rug Pull incidents, shown in Rug-Pull-Incidents.xlsx.The complete code will be open-sourced after our paper is accepted.Prerequisites:gigahorse-toolchain: refer to the nevillegrech/gigahorse-toolchain for setup details.Python >= 3.8Networkx 3.1, Pytorch 2.4.0, torch_geometric 2.5.3, DGL 2.4.0Datasets:Incidents-Source-Code.zip: The source code of collected Rug Pull incidentsNormal-Bytecode.zip: The bytecode of our benign tokensRug-Bytecode.zip: The bytecode of out Rug Pull tokensCode:Flow-Analysis：The implementation of Flow Analysis based on Gigahorse toolRPhunter: The implementation of Graph Construction (SRCG and TFBG) and Detect Model based on PythonExperiments:SOTA: The experiment results of SOTA methods on our ground-truth datasetReal-world Rug Pull crypto tokens: The experiment results of our method on the real-world Ethereum Mainnet