SCMDetector: Smart Contract Malicious Detection Technique based on GLM and ABLSTM-A

Abstract

Existing static detection methods often fail to cap-ture dynamic interactions in smart contracts, resulting in low detection accuracy. Noise from irrelevant data can also affect the precision of vulnerability detection. This paper introduces a new method for detecting malicious smart contracts-GLM-ABLSTM-A, which integrates a General Language Model (GLM) with an Attention-based Long Short-Term Memory (ABLSTM) network. The method aims to address the limitations of static detection techniques, such as low accuracy and limited practicality, focusing on the interactivity and collaboration of smart contract systems. It compiles malicious contract code into Java and labels it, then preprocesses the code with GLM to ex-tract relevant textual information, reducing noise in the detection process. Finally, the extracted feature vectors are fed into the ABLSTM-A classifier. This technique introduces a feature extraction framework based on GLM, combined with the ABLSTM-A classifier, which enhances both the accuracy and efficiency of malicious contract detection and improves the in-teractivity and adaptability of the detection system.