Design and evaluation of highly accurate smart contract code vulnerability detection framework

Smart contracts are self-executing programs stored and executed on a blockchain platform. However, previous studies demonstrated that developing secure smart contracts is not easy. Unfortunately, the use of insecure smart contracts results in a significant financial loss for service providers or customers. Therefore, identifying security vulnerabilities in smart contracts would be essential in blockchain platforms using smart contracts. In this paper, we present SmartConDetect as a tool for detecting security vulnerabilities in Solidity smart contracts. SmartConDetect is a static analysis tool that extracts code fragments from Solidity smart contracts and uses a pre-trained BERT model to find susceptible code patterns. To demonstrate the performance of SmartConDetect, we use two public datasets, and our dataset (SmartConDataset) collected from the real-world Ethereum blockchain network. Our experimental results show that SmartConDetect significantly outperforms all state-of-the-art methods, achieving 90.9% F1-score when using our own dataset. Specifically, SmartConDetect is about 2 times faster than SmartCheck in detection. Furthermore, we conduct a real-world case study to analyze the distribution of detected vulnerabilities.