基于深度学习的智能合约漏洞检测方法综述

A survey of smart contract vulnerability detection methods based on deep learning

摘要:

智能合约是区块链三大特点之一，也是区块链具有应用价值和灵活性的领域.本质上，智能合约是一段用特定脚本语言实现的代码，不可避免地存在安全漏洞风险.如何及时准确地检查出各种智能合约的漏洞，就成为区块链安全研究的重点和热点.为了检测智能合约漏洞，研究者提出了各种分析方法，包括符号执行、形式化验证和模糊测试等.随着人工智能技术的快速发展，越来越多基于深度学习的方法被提出，并且在多个研究领域取得了很好的效果.目前，针对基于深度学习的智能合约漏洞检测方法并没有被详细地调查和分析.本文首先简要介绍了智能合约的概念以及智能合约漏洞相关的安全事件；然后对基于深度学习的方法中常用的智能合约特征进行分析；同时对智能合约漏洞检测中常用的深度学习模型进行描述.此外，为了进一步推动基于深度学习的智能合约漏洞检测方法的研究，本文将近年来基于深度学习的智能合约漏洞检测方法根据其特征提取形式进行了总结分类，从文本处理、静态分析和图像处理3个角度进行了分析介绍；最后，总结了该领域面临的挑战和未来的研究方向.

Abstract:

Smart contracts are one of the three major characteristics of blockchain, and they are also areas where blockchain has application value and flexibility. In essence, a smart contract is a piece of code implemented in a specific scripting language, which inevitably has the risk of security vulnerabilities. How to accurately and timely detect the vulnerabilities of various smart contracts has become the focus and hot spot of blockchain security research. To detect vulnerabilities in smart contracts, researchers have proposed various analysis methods, including symbolic execution, formal verification and fuzzing. With the rapid development of artificial intelligence technology, more and more deep learning-based methods have been proposed and have achieved good results in several research areas. At present, deep learning-based smart contract vulnerability detection methods have not been investigated and analyzed in detail. This paper first briefly introduces the concept of smart contracts and security events related to smart contract vulnerabilities, then introduces the commonly used smart contract features in deep learning-based methods, and describes the deep learning models commonly used in smart contract vulnerability detection. In addition, in order to further promote the research of deep learning-based smart contract vulnerability detection methods, the recent deep learning-based smart contract vulnerability detection methods are summarized and classified according to their feature extraction forms, and are analyzed and introduced from three perspectives: text processing, static analysis and image processing. Finally, the challenges and future research directions in this field are summarized.