Account Profile And Anomaly Detection Based On Blockchain Financial Transaction Behavior

In recent years,blockchain-based cryptocurrencies have attracted a lot of attention,especially popular blockchain applications such as Ethereum.A large number of financial crimes have occurred in blockchain application,which cause a negative impact on the financial security of the blockchain ecosystem and the development of blockchain technology.Duo to the anonymity of blockchain,blockchain supervision is facing huge challenge.With the popularization and importance of blockchain technology,it is important to conduct research on financial supervision of blockchain.To this end,based on Ethereum blockchain platform,this paper takes the anonymous accounts in transfer transaction network as the research object and conducts research in the field of account information mining for account profiling of anonymous accounts.At the same time,research on abnormal account detection based on account profiles,discuss and solve real-world data imbalance problem to improve the performance of abnormal account detection.The contributions of this paper mainly include:(1)Extract account transaction behavior features based on the three types of transaction behavior of Ethereum transfer transaction,contract creation transaction and contract invocation transaction.Conduct comprehensive analysis of all financial transaction behaviors in Ethereum.At the same time,in order to explore the potential relationship between the account in the transfer transaction network and the neighbor account in the above three transactions,this paper extracts the cascading features based on account transaction behavior features and the cascading feature extraction method,and provides rich feature set for constructing account profile.And experiments have verified its effectiveness in abnormal account detection.(2)In order to describe the importance of accounts and further build account profile,this paper proposes an online Time Adaptive Weighted Katz centrality measurement algorithm and optimizes space storage in combination with Time Adaptive Sketch.The experiments show that the applicability of the centrality algorithm in Ethereum transaction network and the efficiency of space storage.(3)Based on the above-mentioned centrality measurement algorithm,this paper proposes Weighted Stream Walk algorithm for Ethereum transaction network,an online graph embedding algorithm that combines transaction time and amount information to mine the spatial structure information of the transaction network for further constructing deep account profile.So that the similarity between accounts with more frequent transactions and larger transaction amounts is higher.The experimental results show the effectiveness of this deep account profile.At the same time,combining the account transaction behavior features and the cascade features can further improve the quality of the account features.(4)Based on the account profiling technology proposed in this paper,this paper conducts research on abnormal account detection and feature selection.From the perspectives of improving the quality of account characteristics and solving the problem of data imbalance,gradually improve the performance of abnormal account detection to an excellent level.Finally,this paper selects the Light GBM model and the Self-paced Ensemble data imbalance integrated learning algorithm as the abnormal account detection model.At the end of this article,a prototype system is designed and implemented,which makes it possible to implement efficient financial supervision of Ethereum.