Exploring the potential of large-language models (LLMs) for student feedback sentiment analysis

Large-language models (LLMs) have demonstrated remarkable performance across a wide range of natural language processing (NLP) tasks, including synthetic text generation, classification, question answering, and language translation. In this paper, we explore the potential of leveraging these LLMs for sentiment analysis or opinion mining of students’ feedback about their teachers, typically collected at the end of a course. Analyzing students’ sentiments is crucial for academic decision-making. We conducted our study by employing ChatGPT, a popular LLM, to perform sentiment classification on a diverse dataset of student feedback. This dataset was collected and scientifically labeled with sentiment annotations by our experienced annotators team. Our findings demonstrate the immense promise of using LLMs in accurately classifying students’ feedback into positive, negative, or neutral sentiments. The ChatGPT model achieved an impressive overall F1-score of 88%, outperforming state-of-the-art deep learning and transformer-based models. These results show the significance of LLMs in advancing sentiment analysis in educational contexts and provide valuable insights for educators and administrators to enhance the learning experience.