Research on Log Anomaly Detection Based on Sentence-BERT

Log anomaly detection is crucial for computer systems. By analyzing and processing the logs generated by a system, abnormal events or potential problems in system can be identified, which helpful its stability reliability. At present, due to expansion of scale complexity software systems, amount log data grows enormously, traditional methods have been unable detect anomalies time. Therefore, it important design with high accuracy strong generalization. In this paper, we propose method LogADSBERT, based on Sentence-BERT. This adopts Sentence-BERT model extract semantic behavior characteristics implements through bidirectional recurrent neural network, Bi-LSTM. Experiments open set show that LogADSBERT better than existing methods. Moreover, robust even under scenario new event injections.