Predicting software developer sentiment on self-admitted technical debt

Technical debt is a metaphor for sacrificing long-term code quality in order to achieve short-term project goals. The technical debt that developers intentionally introduce into project is called self-admitted technical debt (SATD), which usually exists in the form of code comments in software projects. The existence of SATD poses a huge challenge to software quality and robustness. Analyzing the sentiments of SATD helps to understand the behavioral habits of developers when annotating SATD. In order to explore the performance of generative pre-trained models in SATD sentiment prediction, a SATD sentiment prediction method based on the GPT-3.5-turbo fine-tuning model is proposed, and research is carried out on 18 open source projects. Empirical results show that, compared with a set of traditional machine learning and deep learning techniques, the fine-tuning GPT-3.5-turbo model improves the evaluation indicators precision, recall and F1-score by 14.2%, 11.5%, and 17.3%, respectively. The effectiveness of the SATD sentiment prediction method based on GPT-3.5-turbo is verified, indicating the potential of generative pre-trained models to capture nuanced complex sentiment— i.e ., developer comments that combine neutral technical observations with subtle negative cues such as hesitation or frustration. However, our comparison does not include recent large language model (LLM)-based approaches, which are reserved for future investigation.