摘要
针对现有的基于深度学习智能合约漏洞检测方法无法有效利用上下文信息,提出一种基于异构图的智能合约漏洞检测方法。通过将合约源码解析为包含数据流和控制流的符号图,然后使用图神经网络对图进行表征学习,并通过神经网络进行漏洞预测。在ESC和VSC两个数据集上进行实验,和现有工具以及模型进行对比,结果表明该方法在准确率、召回率、精度、F1分数4个指标均取得提升。
To address that the existing smart contract vulnerability detection-based deep learning cannot effectively use context information,This paper proposes a smart contract vulnerability detection based on a heterogeneous graph,Which parses the contract into a Symbol diagram containing data-flow edge and control-flow edge.Then it uses graph neural net-works to perform representation learning on the graph,finally,the vulnerability prediction is performed through the neural networks.Experiments are conducted on ESC and VSC data sets,and comparing them with existing tools and models,the results show that the method has improved in the four indicators of accuracy,recall,precision,and F1-score.
作者
侯羿杉
王燚
HOU Yishan;WANG Yi(College of Cybersecurity,Chengdu University of Information Technology,Chengdu 610225,Chin)
出处
《成都信息工程大学学报》
2025年第1期7-13,共7页
Journal of Chengdu University of Information Technology
基金
四川省科技计划资助项目(2023DYF0380、2021ZYD0011)
国家社会科学基金资助项目(23BSH061)。
关键词
智能合约
漏洞检测
深度学习
图神经网络
smart contract
vulnerability detection
deep learning
graph neural network
