12-llmloganalyzer

LLMLogAnalyzer - Clustering-Based Log Analysis Chatbot

• Paper: https://arxiv.org/abs/2510.24031
• Status: Read

Takeaway

• A chatbot for log analysis, not a log parser — different goal from LAPP Phase 1
• Pipeline: router (classify user query) → log recognizer (identify log type) → log parser (cluster + extract) → LLM generates answer
• Solves LLM context window problem by clustering logs first, then feeding summaries instead of raw logs
• Tested on 4 log domains: beats ChatGPT/ChatPDF/NotebookLM by 39-68% on summarization/pattern/anomaly tasks
• 93% less output variability (more consistent answers)
• For LAPP: the “cluster first, then summarize” idea is relevant for Phase 2 reporting, but the chatbot interface itself is not our focus