Automating Code-Change Rationale Generation to Support Software Maintenance Tasks

Mehedi Sun. Automating Code-Change Rationale Generation to Support Software Maintenance Tasks. In the International Conference on Software Engineering (ICSE) 2026 Doctoral Symposium.

Abstract

Understanding the rationale behind code changes is essential for software maintenance, yet that information is often fragmented across artifacts and difficult to use. This research investigates fine-grained, personalized, and task-specific explanations of why changes occur. It develops techniques to extract and generate multi-document rationale summaries, models developer profiles to tailor explanations to their expertise and familiarity, and generates task-specific rationale for code review, debugging, and onboarding. The work contributes empirical insights, a cross-artifact rationale benchmark, and an automated approach for personalized, task-aware rationale generation to support developer understanding and software maintenance.