Log Anomaly Detection with LINE

To detect anomalies in large console log series, we propose a novel clustering method with graph-embedding. We leverage both the semantic knowledge and relationship among individual log messages to create meaningful and robust embedding for each event. The embedding is then aggregated to form sequence feature vectors for log anomaly detection through clustering and distance weighing. With the employment of efficient clustering assignment model, we reduce the offline computation time and complexity while preserving the distinct identification of anomalies.

The code is published in the repo and the report paper can be accessed through the pdf link.