Anomaly Detection from System Logs based on Deep Learning Techniques

Modern computer systems are mostly distributed and have a complex and large structure. Therefore, it's important that the system operates safely and effectively. Fortunately logs record the state of system operation, so logs play an important role in system anomaly detection. It is a proven method to use deep learning methods for system anomaly detection. In this paper, we improve the log parsing method and use the Long Short-Term Memory neural network in combination with the Transformer deep neural network to model the log data into log sequences for anomaly detection. The experimental results show that the improved log parsing method is significantly more efficient and the F1 scores of this paper's model are significantly better than other models.