A Comparative Study of Bug Triage Representation and Classification Approaches from Canonical to Large Language Models

Bug triage is the task of assigning newly reported bugs to the proper developers or team for resolution. This is a critical point in software maintenance as it directly influences the time and correct allocation that impact the efficiency and effectiveness of the software process. In a global aspect, the number of teams/developers is extensive, which brings a challenge for bug triage. Traditional approaches to assign bugs struggle with the complexity of the problem. This paper proposes a comparative assessment for different text representation combined with text classification approaches for automated bug report triage by incorporating Large Language Models (LLMs) into the classification pipeline to surpass the limitations of canonical methods. Traditional classification methods were compared with LLM-enhanced models across accuracy metric. The results demonstrate an improvement in triage accuracy when utilizing the fine-tuned LLM, highlighting their potential to provide developer-appropriate bug assignments.