Research On Key Techniques For Identifying Abnormal Transaction Behaviors In Blockchain

With the development of blockchain technology,blockchain is often used in illegal financial activities due to its decentralized and difficult to regulate characteristics.The use of technical means to detect abnormal transaction behavior in blockchain,monitor and manage abnormal transaction behavior in blockchain,is of great significance for protecting the ecosystem of blockchain systems,ensuring financial and social security.At present,there are three problems in the identification method of blockchain abnormal transaction behavior: the first is that the identification target of abnormal transaction behavior caused by blockchain anonymity is not clear,and the current algorithm intermediate process is not traceable,and the algorithm operation efficiency is low due to repetitive work.The second aspect is the difficulty in accurately characterizing and predicting blockchain trading behavior,and the low computational efficiency of graph methods in massive transaction data.The third aspect is the recognition method for abnormal transaction behavior in blockchain,which considers issues such as a single dimension and insufficient accuracy in identifying scenarios such as blockchain transaction pool accumulation.In order to solve the above three problems,this thesis studies a more efficient and accurate method for identifying abnormal transaction behavior in blockchain.The main research content and innovative points are as follows:1)Aiming at the problem that the identification target of abnormal transaction behavior caused by anonymization of blockchain transaction entity identity and fragmentation of transaction information is not clear,this thesis researches the incremental identification method of blockchain transaction entity based on heuristic address clustering.In the UTXO ledger model,how to divide transaction addresses into transaction entities is the basis for identifying abnormal transaction behavior in blockchain.This section proposes three heuristic methods to improve existing blockchain address clustering algorithms based on association and aggregation methods,including studying address clustering optimization methods such as zero finding addresses and untrusted transactions.At the same time,in response to the problems of untraceable intermediate processes and low algorithm efficiency caused by repetitive work in the Bitcoin address clustering method,this section studies the Bitcoin address incremental clustering method based on transaction entities to solve the problem of inability to store and analyze the intermediate processes of address clustering.2)In response to the problem of low processing efficiency of existing blockchain transaction behavior prediction methods based on transaction graphs in massive data,this thesis takes the example of Taifang blockchain to study a blockchain transaction behavior prediction method based on transaction subgraph partitioning.By extracting transaction relationships from Ethereum block data,we constructed a transaction graph with transaction addresses as nodes and transaction relationships as edges in this section,and proposed a method for constructing transaction subgraphs that can be divided into transaction subgraphs based on connectivity.Addressing the difficulties and low accuracy of feature extraction.On the basis of the transaction subgraph,we propose a GNN based random walk algorithm in this section,which can consider transaction amount,transaction time,and other factors.In order to improve the accuracy of blockchain transaction behavior prediction,we propose a node connection relationship measurement method in this section,and verify the effectiveness of this method through real transaction data experiments.3)Based on the objectives and behavior patterns of identifying abnormal blockchain trading behavior,we propose a multi-dimensional method for identifying abnormal blockchain trading behavior in this section.In the identification scheme of abnormal trading behavior in the transaction itself,we determine the purpose of the transaction itself by verifying whether the transaction purpose is digital currency transfer,and then speculate whether there is abnormal trading behavior in it.We constructed a transaction autonomous message analysis system based on deep learning models by designing methods for extracting and preprocessing blockchain transaction autonomous messages,and ultimately formed a Bitcoin transaction autonomous message dataset.In the identification scheme of abnormal transaction behavior between transactions on the chain,in addition to considering the traditional transaction correlation,we also proposed an identification method of abnormal transaction behavior that takes into account the transaction time dependency.This method can identify abnormal transaction behavior that requires a large number of transactions,such as currency mixing,with a high accuracy rate.By designing and implementing a blockchain transaction Memory pool monitoring algorithm,we can identify abnormal transaction behaviors when blockchain transaction pools are stacked.In addition,we also propose a visual analysis based method for identifying abnormal trading behaviors in this section,which can convert transactions with time dependencies from list form to chart form,making it more intuitive to determine the abnormal trading behaviors within them.