부산대학교 도서관

Technical debt is a pervasive issue in software development, often arising from trade-offs made during development, which can impede software maintainability and hinder future development efforts. Self-admitted technical debt (SATD) refers to instances where developers explicitly acknowledge suboptimal code quality or design flaws in the codebase. Automated detection of SATD has emerged as a critical area of research, aiming to assist developers in identifying and addressing technical debt efficiently. However, the enormous variety of feature extraction approaches of NLP and algorithms employed in the literature often hinder researchers from trying to improve their performance. In light of this, this systematic literature review proposes a taxonomy of feature extraction techniques and ML/DL algorithms used in technical debt detection: its objective is to compare and benchmark their performance in the examined studies. We selected 53 articles that passed the quality evaluation of the systematic review. We then investigated in depth which feature extractions and algorithms were employed to identify technical debt in each software development activity. All approaches proposed in the analyzed studies were grouped into NLP, NLP+ML, and NLP+DL. This allows us to discuss the performance in three different ways. Overall, NLP+DL group consistently outperforms in precision and F1-score for all projects, and in all but one project for the recall metric. Regarding the feature extraction techniques, the PTE consistently achieves higher precision, recall, and F1-score for each project analyzed. Furthermore, TD types have been mappep to software development activities; this served to determine the best-performing feature extractions and algorithms for each development activity. Finally, based on the review results, we also identify implications that could be of concern to researchers and practitioners.