ABSTRACT With the rapid development of Internet technologies, network security threats have significantly increased, emphasizing the need for accurate and efficient network intrusion detection systems (IDS). While traditional IDS methods rely on manual feature selection or deep learning, these approaches often fail to capture the full complexity of network data, especially the diverse structural and content‐related features. To address these limitations, this paper proposes a novel self‐supervised contrastive learning algorithm for network intrusion detection based on a graph isomorphism network ( SCGIN‐ID ). The method transforms network traffic into graph representations, converting the task into a node classification problem. By sampling positive and negative subgraphs centered on target nodes, the proposed approach leverages graph isomorphism networks to capture both structural and content‐based anomalies through self‐supervised contrastive learning. Extensive experiments on two benchmark datasets demonstrate that SCGIN‐ID achieves superior performance compared to five representative baseline models, providing an effective solution for comprehensive and accurate network intrusion detection.

Network Intrusion Detection Systems (NIDSs) play a crucial role in safeguarding the security of protected computer networks. Although numerous machine learning algorithms, especially deep learning algorithms, have achieved remarkable results, their generalization ability is limited due to the following critical challenges. First, most of existing methods heavily rely on the handcrafted features extracted from packets or network flows. Second, few studies have been devoted to adaptively highlighting the characteristics of certain traffic features and thus extracting discriminative representations from input network data. In this paper, we propose a Multi-level ATTention-enhanced rEpresentation leaRning model (MATTER) to address the aforementioned challenges. Specifically, a multi-scale Convolutional Neural Network (CNN) is employed to extracted representations from the raw packet content of a network flow. Then, a multi-level attention module with spatial, channel and temporal attention mechanisms is leveraged to enhance the discrimination of the extracted features. Extensive experiments on two benchmark datasets demonstrate that our proposed MATTER is superior to other state-of-the-art approaches in terms of both accuracy and F1 score.