信息安全研究 ›› 2023, Vol. 9 ›› Issue (1): 13-.

• 学术论文 • 上一篇    下一篇

基于注意力机制的图神经网络加密流量分类研究

喻晓伟;陈丹伟;   

  1. (南京邮电大学计算机学院、软件学院、网络空间安全学院南京210023)
  • 出版日期:2023-01-01 发布日期:2022-12-30
  • 通讯作者: 喻晓伟 硕士研究生.主要研究方向为网络安全、加密流量分类. yuxw129@qq.com
  • 作者简介:喻晓伟 硕士研究生.主要研究方向为网络安全、加密流量分类. yuxw129@qq.com 陈丹伟 博士,副教授.主要研究方向为信息安全、计算机应用. chendw@njupt.edu.cn

Research on Encrypted Traffic Classification of Graph Neural Network  Based on Attention Mechanism

  • Online:2023-01-01 Published:2022-12-30

摘要: 对于加密流量的精确化识别,现有的基于机器学习和基于图的解决方案需要人工进行特征选择或者精度较低.使用一种基于图神经网络的加密流量识别方法,通过将网络流量数据转换为图数据,保留了网络数据流的丰富表示,将网络流量分类问题转换为图分类问题.并设计了一个基于自注意力机制的图分类模型进行加密流量的分类.实验结果表明,该方法对基于安全套接层(secure socket layer, SSL)的虚拟专用网(virtual private network, VPN)加密流量具有较好的分类效果,分类准确率有较大提高.

关键词: 加密流量分类, 流量图, 注意力机制, 图神经网络, SSL加密流量

Abstract: For precise identification of encrypted traffic, existing machine learningbased and graphbased solutions require manual feature selection or have low accuracy. Using a graph neural networkbased encrypted traffic identification method, the network traffic classification problem is transformed into a graph classification problem by converting the network traffic data into graph data, preserving the rich representation of the network data flow. And this paper designs a graph classification model based on selfattention mechanism to classify encrypted traffic. The experimental results show that the method has a good classification effect on the encrypted traffic of Virtual Private Network (VPN) based on Secure Socket Layer (SSL), and the classification accuracy is greatly improved.Key words

Key words: encrypted traffic classification, traffic graph, attention mechanism, graph neural network, SSL encrypted traffic