GLM-Rf: Malicious smart contract static detection scheme based on GLM and Random forest

This paper introduces an innovative approach to improve the detection of malicious smart contracts, presenting the GLM-Rf (Malicious Smart Contract Detection Scheme Based on General Language Model and Random Forest). The scheme places a primary emphasis on the interactive and collaborative nature of smart contract systems, aiming to address prevalent issues such as limited detection accuracy and practicality associated with static detection methods. GLM-Rf initiates by preprocessing malicious contract code, extracting relevant text from the entire code to mitigate the impact of extraneous information on detection results. Furthermore, it employs the WordPiece word segmentation algorithm to segment the smart contract code. Subsequently, the preprocessed code undergoes feature extraction through a network based on the General Language Model (GLM), generating feature vectors that encapsulate high-level semantics. These feature vectors are then input into a carefully designed random forest classifier, significantly enhancing the efficiency and precision of vulnerability detection. GLM-Rf represents a promising breakthrough in the field of malicious smart contract detection, offering a more robust and adaptable solution to current challenges, with a strong focus on interactivity.