Implementing and Evaluating Automated Bug Triage in Industrial Projects

Resolving bugs on time is essential for software development and is critical in industrial projects because it directly affects businesses. Automatic bug triage has been investigated to increase software productivity, and research has become more active as machine learning techniques have improved. However, most research has focused on open-source projects, whereas studies on industrial projects remain limited. The research gap in previous studies is that the research has directly triaged developers, reducing accuracy in industrial projects where organizational structures frequently change. Moreover, developers often move between teams, making this approach less effective. The research in this article applies automatic bug triage to industrial projects by adapting the characteristics of industrial projects. Addressing these limitations establishes an approach that is better suited to industrial projects and has enhanced accuracy. Based on this background, we propose a novel approach to triage developers associated with component-based developer lists. Each component has an associated list of developers, and the triage results of the model are limited to selecting from among the listed developers, enhancing triage accuracy. The proposed approach reflects the characteristics of industrial projects and addresses the dynamic workload adjustments in a component-based team structure. The proposed approach improves the results by 6.2 percentage points over human triage for top-1 results, suggesting that this approach could be further expanded for broader application in industrial contexts. Future research should focus on refining the proposed method with real-time feedback and experiment with a broader dataset for generalizability and scalability.