Scholarlink

<p>Protecting networks from malicious access is a highly challenging task. Identifying and detecting anomalous paths can effectively filter out potential network threats. Despite extensive research in this field, the advent of new network technologies and the increase in connected devices have led to more diverse network attacks. Traditional anomaly detection methods struggle to capture complex relationships between nodes. Graph neural networks (GNNs), with their multi-layer structures, can efficiently extract multi-attribute features of nodes. We propose an efficient method for anomalous path detection. By incorporating a multi-head attention mechanism into GNNs, we dynamically adjust node weights to capture complex dependencies. To accurately identify anomalous nodes, we also designed a new threat evaluation scoring formula, enhancing the robustness and accuracy of anomaly detection. Experimental results on the CIC-IDS-2017 dataset show that our model achieved nearly 100% accuracy and the fastest response time. Compared with previous work, our model provides faster decision-making and significantly improved evaluation performance. Our research demonstrates the potential of GNNs and multi-head attention mechanisms in network security, offering strong support for addressing increasingly complex network threats.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Network Security Anomaly Node Detection Based on Graph Neural Network and Attention Mechanism

  • 期刊论文

  • 发表日期：

    2025年7月7日

查看全文
  • Deng Zhang,
  • Juan Wang,
  • Hanjun Gao,
  • Zhou Wei Ni,
  • Hao Zhang

摘要

Protecting networks from malicious access is a highly challenging task. Identifying and detecting anomalous paths can effectively filter out potential network threats. Despite extensive research in this field, the advent of new network technologies and the increase in connected devices have led to more diverse network attacks. Traditional anomaly detection methods struggle to capture complex relationships between nodes. Graph neural networks (GNNs), with their multi-layer structures, can efficiently extract multi-attribute features of nodes. We propose an efficient method for anomalous path detection. By incorporating a multi-head attention mechanism into GNNs, we dynamically adjust node weights to capture complex dependencies. To accurately identify anomalous nodes, we also designed a new threat evaluation scoring formula, enhancing the robustness and accuracy of anomaly detection. Experimental results on the CIC-IDS-2017 dataset show that our model achieved nearly 100% accuracy and the fastest response time. Compared with previous work, our model provides faster decision-making and significantly improved evaluation performance. Our research demonstrates the potential of GNNs and multi-head attention mechanisms in network security, offering strong support for addressing increasingly complex network threats.