Bilkent University
Department of Computer Engineering
CS 590/690 SEMINAR
Enhancing Software Maintainability via Inline Comment Repair During Code Reviews
Hatice Kübra Çağlar
Master Student
(Supervisor: Asst.Prof.Eray Tüzün)
Computer Engineering Department
Bilkent University
Abstract: Code comments are crucial to software documentation, supporting code comprehension and long-term maintenance. In modern collaborative development, especially on platforms like GitHub, pull requests (PRs) are primary vehicles for code updates and reviews. However, inline comments submitted with PRs might suffer from quality issues, in other words, smells. Addressing these smells early in the development process can significantly improve software maintainability and reduce technical debt over time. Our research aims to improve the quality of inline code documentation in pull requests by automatically detecting and repairing comment smells. By targeting the PR stage, we seek to assist developers in submitting cleaner, more helpful comments before they are merged into the codebase. We developed a comment smell detection and repair tool that activates when a pull request is submitted. The tool uses a labeled dataset of code-comment pairs and leverages large language models (LLMs) to detect common types of comment smells. Once identified, the tool suggests improved versions of problematic comments or generates appropriate comments when they are missing. These suggestions are evaluated both quantitatively—using precision, recall, accuracy, and F1 metrics for smell detection performance and BLEU, ROGUE, and METEOR for repair performance—and qualitatively through expert reviews.
DATE: April 14, Monday @ 13:30 Place: EA 502