Unknown Vulnerability Detection in Smart Contracts with Multi-Class Model using C-LSTM

Smart contracts have enhanced the blockchain ecosystem, while they have also greatly increased the risk of attack due to their numerous security vulnerabilities. Traditional detection techniques and static analysis methods mainly focus on detecting known vulnerabilities. Aiming at unknown vulnerabilities, we propose a novel detection method and build a multi-class model with C-LSTM (Convolutional Neural Network, Long Short-Term Memory), combining opcode sequences and deep learning. The proposed unknown vulnerabilities detection method leverages the similarity between opcode sequences of unknown and partially known vulnerabilities. It supports a faster response to a large number of contracts that need to be tested, but also can discover unknown vulnerabilities in advance. We conduct trials on 14,150 opcode sequences generated by executing transactions on Ethereum. Extensive experiments have verified the effectiveness and efficiency of the model.