이상적인 로그 템플릿 추출을 위한 NER의 활용 가능성 탐구

초록

Manual monitoring of a system is becoming difficult due to increasing amount of log messages. Accordingly, new log analysis techniques using data mining models are introduced, However most models require structured input. Therefore, a technique that can convert unstructured log messages into structured log templates is required. Existing clustering-based techniques suffer from an inability to detect entities made of multiple tokens. To address this problem, this paper explored the feasibility of extracting log templates using named entity recognition (NER). Specifically, through experiments, we verify the feasibility of NER to overcome the limitations of traditional rule-based parsing methods. The results demonstrate that NER can meaningfully identify log entities, achieving an F1-score of 0.902.

키워드