Research And Implementation Of Ethereum Fraud Detection Technology Based On Deep Learning

Blockchain,which originated from the famous Bitcoin digital currency,is a decentralized,distributed and trusted database maintained by a peer-to-peer network through a specified consensus mechanism and is commonly used in the field of digital asset cryptocurrencies.Blockchain technology itself does not support complex logical operations and is less scalable.Ethereum is a blockchain platform that supports smart contracts,which can build many automated applications through smart contracts,increasing the scalability of the blockchain system.Smart contracts have greatly driven the widespread use of blockchain technology in areas such as healthcare,finance and the Internet of Things,but the imperfection of blockchain systems has led to a high incidence of criminal activity.For example,frauds such as phishing and Ponzi schemes in Ethereum have seriously threatened the security of users’ properties.Therefore,there is an urgent need for a detection method that can efficiently and accurately identify these fraudulent scams.Currently,most of the research related to blockchain fraud detection is based on machine learning related methods,but most of these detection methods suffer from insufficient data sources and imperfect algorithms and models.Some researchers identify fraudulent accounts by extracting the bytecode frequency features of smart contracts.This method does not directly rely on smart contract code,which greatly reduces the complexity of data feature extraction,but this detection method is not applicable to fraudulent scams without smart contract participation.To solve the above problems,this thesis designs detection schemes for phishing scams and Ponzi schemes based on their design differences,respectively,and constructs a graph dataset based on transaction records for the training of the model.The main research elements of this thesis are as follows:(1)The attack principles of Ponzi scheme and phishing scam on Ethereum are analyzed,and their design differences are derived by comparing these two fraud attack principles,which provide ideas for the data feature construction and model design of the fraud detection scheme in this thesis.(2)Based on the design differences of these two fraudulent scams,different feature datasets are constructed for model training of phishing fraud detection model and Ponzi scheme detection model,respectively.(3)The Edge-sampling To Node Vector algorithm(Esmp2Nvec)is designed for the generation of initial embedding features of transaction graph vertices,which uses the edge features of the transaction graph to generate embedding vectors for the nodes and adds the directional features of the transactions in the process of generation of the node embedding vectors.(4)A graph classification network model is proposed for Ethereum phishing scam detection.In this thesis,based on the feature of transaction graph data in phishing fraud,Trans Detection Net,a graph classification network model for phishing fraud detection,is constructed based on Esmp2 Nvec algorithm combined with GCN and Graph SAGE to achieve deep extraction of transaction graph features.(5)For Ponzi schemes that rely on smart contracts,this thesis construct SGCSNet,a multi-channel graph classification network model for Ponzi scheme detection,to extract the features of the transaction graph based on the features of its transaction graph;in addition,this thesis perform multimodal feature fusion of the features of the transaction graph with the features of the smart contract’s opcode so that the data from each perspective collaboratively represent the features of the contract account.Experimental results on real data on Ethereum show that our method achieves good results in both phishing account identification and Ponzi scheme identification.