Research On Ponzi Scheme Detection Of Smart Contract Based On Graph Neural Network

With the rapid development of blockchain technology,the application of Ethereum has become increasingly widespread.Developers have deployed many smart contracts in Ethereum to achieve various functions.Unfortunately,speculators introduced Ponzi schemes in the traditional financial sector into smart contracts,causing millions of dollars in losses to investors.Currently,in the context of Internet finance,there are few quantitative identification methods for new fraud patterns and even fewer Ponzi scheme detection methods in Ethereum.Therefore,how to effectively detect whether a smart contract is a Ponzi scheme is crucial to the development of Ethereum.The existing Ponzi scheme detection methods are divided into rule-based detection and machine learning-based detection methods.The former cannot identify more complex Ponzi schemes due to the incompleteness of the rules and has a higher false alarm rate;The latter simply uses the frequency of the opcode as the detection feature,and its incomplete semantic expression of the smart contract makes the detection effect limited.To solve the problems of the above methods and further improve the effect of Ponzi scheme detection,this thesis proposes a Ponzi scheme detection method based on graph neural network.This method fully retains the structure and semantic information of smart contracts through the data structure of the graph and detects whether the smart contract is a Ponzi scheme by training the graph neural network model.The main work of this thesis is as follows:(1)This thesis analyzes the existing Ponzi scheme detection mechanisms,and introduces in detail the basic principles,implementation process,and existing problems of each detection method.This thesis summarizes the disadvantages and deficiencies of the current Ponzi scheme detection methods such as incomplete rule statistics and incomplete preservation of the semantic information and structural information of the contract itself.To solve these problems,this thesis puts forward a method of Ponzi scheme detection based on graph neural network.(2)Aiming at the problems of the lack of semantic information expression of existing Ponzi scheme detection methods and the poor detection effect of complex contracts,this thesis proposes a Ponzi scheme detection scheme based on graph neural network.First,we analyze the characteristics of the bytecode in the smart contract and compose the graph,and then use the embedding algorithm to vectorize the node feature information and edge feature information in the graph.Finally,this thesis uses the processed data to train the graph neural network model and utilizes the trained network model to detect the Ponzi scheme.(3)According to the Ponzi scheme detection method in smart contract based on graph network proposed in this thesis,we design and implement a prototype system.First,we introduce the details of the system.Then we construct a Ponzi scheme dataset and compare our prototype system with the existing Ponzi scheme detection tools to verify the effectiveness.Experimental results show that under the open-source Ponzi scheme dataset,the method proposed in this thesis improves the F1 score of the Ponzi scheme detection method based on machine learning by 6% to 9%.Finally,to verify the effectiveness of the prototype system in this thesis in a real environment,the smart contracts deployed in Ethereum were tested.We found and manually verified 11 Ponzi schemes outside of the open-source dataset.