Research on System Log Anomaly Detection Method Based on Graph Neural Network
Yiming Wang · 2025
System log anomaly detection is crucial to ensure the safety of system operation. The anomaly detection method based on graph neural network constructs a dynamically evolving graph structure through$\log$sequences, extracts node features by using multi-attention mechanism, and calculates anomaly scores by combining with Mahalanobis distance. It achieves 95.8% detection accuracy on HDFS dataset, which is$\mathbf{5. 7 \% - 1 2. 3 \%}$improvement over existing methods. Experiments show that the dynamic graph updating mechanism and the multi-head attention structure significantly improve the model performance, with a 43.2 % improvement in computational efficiency. The method provides a new solution for large-scale syslog anomaly detection.