Smart contract vulnerability detection based on variant LSTM
Ruijie Luo, Luo Feng, Bingsen Wang, Ting Chen · 2022
The smart contracts of Ethereum have brought an essential contribution to the development of blockchain. Nowadays, an increasing number of smart contracts are being deployed on Ethereum, which brings prosperity to Ethereum while also bringing many security risks. According to reports, several attacks due to the smart contract vulnerability have caused huge losses to Ethereum. Therefore, detecting smart contract security vulnerabilities is of great importance. However, the existing work is not sufficient to fully perform this task. For this reason, we propose a variant of the LSTM model to detect smart contract vulnerabilities at the bytecode level. In our variant LSTM model, we interact hidden state of the model with the input sequence multiple times. In this way, the variant model is able to capture more potential features in the bytecode for learning. We used a dataset of 34822 unique smart contracts for training and testing. The results show that the variant LSTM model outperforms the general LSTM model and improves in most metrics.