Sentiment analysis using LLMs: A systematic mapping study in software engineering

Abstract

Assessing programmers' perspectives during software life cycle activities is vital for understanding the reasoning behind software development decisions. Sentiment Analysis (SA) plays a key role in this assessment. Researchers have increasingly leveraged Natural Language Processing (NLP) to conduct such analyses. However, there is a lack of comprehensive and up-to-date studies that explore the challenges and opportunities of applying Large Language Models (LLMs) for SA in Software Engineering (SE). This study systematically maps the topic by reviewing studies published between 2019 and 2025. Following the established protocol, we selected 21 studies that offer a broad perspective on how NLP capabilities within LLMs facilitate SA in software development tasks. Evidence from the selected studies highlights the positive impact of NLP and LLMs on SA in various areas of SE. For instance, some studies apply LLM-driven SA to bug reports and software requirements. Additionally, the mapping study identifies different approaches to utilizing LLMs for SA in SE, including model comparisons, fine-tuning, and prompt engineering.