Multi-Label Vulnerability Detection of Smart Contracts Based on Bi-LSTM and Attention Mechanism
Shenyi Qian, Haohan Ning, Yaqiong He, Mengqi Chen
Abstract
Smart contracts are decentralized applications running on blockchain platforms and have been widely used in a variety of scenarios in recent years. However, frequent smart contract security incidents have focused more and more attention on their security and reliability, and smart contract vulnerability detection has become an urgent problem in blockchain security. Most of the existing methods rely on fixed rules defined by experts, which have the disadvantages of single detection type, poor scalability, and high false alarm rate. To solve the above problems, this paper proposes a method that combines Bi-LSTM and an attention mechanism for multiple vulnerability detection of smart contract opcodes. First, we preprocessed the data to convert the opcodes into a feature matrix suitable as the input of the neural network and then used the Bi-LSTM model based on the attention mechanism to classify smart contracts with multiple labels. The experimental results show that the model can detect multiple vulnerabilities at the same time, and all evaluation indicators exceeded 85%, which proves the effectiveness of the method proposed in this paper for multiple vulnerability detection tasks in smart contracts.