ImproBR: LLM Pipeline Automates Bug Report Enhancement
Researchers have rolled out ImproBR, a new tool that leverages large language models to improve bug reports by fixing issues like vague or missing details. This innovative system combines a finely-tuned DistilBERT model, heuristic methods, and an LLM analyzer, all guided by GPT-4o mini with tailored prompts and a framework that pulls in information from Minecraft Wiki. In tests conducted on Mojira, ImproBR dramatically raised the structural completeness of reports from 7.9% to an impressive 96.4%. It also more than doubled the rate of executable Steps to Reproduce, jumping from 28.8% to 67.6%, and increased fully reproducible bug reports from just 1 to 13 out of 139 complex cases.
Key facts
- ImproBR is an LLM-based pipeline for bug report improvement.
- It targets missing, incomplete, and ambiguous S2R, OB, and EB sections.
- Uses a hybrid detector with fine-tuned DistilBERT, heuristic analysis, and LLM analyzer.
- Guided by GPT-4o mini with few-shot prompts and RAG pipeline.
- RAG pipeline grounded in Minecraft Wiki domain knowledge.
- Evaluated on Mojira bug tracking system.
- Structural completeness improved from 7.9% to 96.4%.
- Executable S2R proportion increased from 28.8% to 67.6%.
- Fully reproducible reports rose from 1 to 13 out of 139 reports.
Entities
Institutions
- Mojira
- Minecraft Wiki