ReLog: A Novel Method for Log Recognition
Bowen Tian, Zhiwen Wang, Guang Chen, Yuchen Zhang, Hong Zhou · 2023
The diversity of logs leads to difficulties in log anomaly detection. Identifying the source of logs can provide the guidelines for subsequent log anomaly detection. To this end, this paper proposes a method for log recognition based on similarity determination, called ReLog. ReLog first takes the feature codes of the target logs and matches them with the log feature codes in the baseline. Then ReLog uses cosine function to calculate the cosine similarity. Finally ReLog selects the item with the highest similarity and meets the minimum threshold as the matching item, so as to identify the log. In order to verify the effectiveness of ReLog, we tested it on datasets of 15 types of logs including Hadoop, Spark, and Mac. The experimental data shows that the recognition rate of ReLog is 99.14% with a acceptable overhead. In addition, it is also able to identify the logs produced by different versions of the same software with a 99.99% recognition rate.