Vulnerability Smart Contract Detection Model based on CUR Matrix Decomposition

Abstract

Smart contracts, an important component of blockchain technology, have received widespread attention due to their decentralized and trustworthy characteristics. However, the security vulnerabilities of smart contracts pose a serious threat to their reliability, causing huge economic losses to users. Existing analysis tools are used to detect security vulnerabilities in smart contracts. However, due to their excessive reliance on hard rules defined by experts when detecting vulnerabilities in smart contracts, the time to perform the detection significantly increases as the complexity of smart contracts increases. In this study, we developed a novel hybrid machine learning model called Bi-CUR. The Bi-CUR model extracts the feature matrix of smart contract opcodes through Bigram and detects smart contract vulnerability through CUR matrix decomposition. It approximates the original matrix by selecting rows and columns, thereby reducing computational complexity while maintaining the features of the data. Compared with traditional vulnerability detection methods, models based on CUR matrix decomposition showed higher efficiency and accuracy. In addition, the model ensured interpretability, which makes it applicable to different types of smart contract vulnerability detection.