Sentiment Analysis in Software Engineering: Evaluating Generative Pre-trained Transformers

Abstract--Sentiment analysis plays a crucial role in understanding developer interactions, issue resolutions, and p roject dynamics within software engineering (SE). While traditio nal SE-specific sentiment analysis tools have made significant s trides, they often fail to account for the nuanced and context-depen dent language inherent to the domain. This study systematically evaluates the performance of bidirectional transformers, such as BERT, against generative pre-trained transformers, speci fically GPT -4o-mini, in SE sentiment analysis. Th e results reveal that fine-tuned GPT -4o-mini performs comparab le to BERT and other bidirectional models on structured and balan ced datasets like GitHub and Jira, achieving macro-averaged F1 - scores of 0.93 and 0.98, respectively. However, on linguist ically complex datasets with imbalanced sentiment distributions, such as Stack Overflow, the default GPT -4o-mini model exhibits superior generalization, achieving an accuracy of 85.3% co m-pared to the fine-tuned model's 13.1%. The study underscores the importance of aligning model architectures with dataset characterist ics to optimize performance and proposes directions for future re search in refining sentiment analysis tools tailored to the SE domai n. Sentiment analysis, a critical subfield of natural language processing (NLP), involves classifying text into sentimen t polarities, such as positive, neutral, and negative. It has been widely studied across various domains, including software engineering (SE), where analyzing sentiments expressed in textual artifacts provides insights into developer intera ctions, issue resolution, and project dynamics.