Search for: All records

The advancement of generative AI, involving the utilization of large language models (LLMs) like ChatGPT to assess public opinion and sentiment, has become increasingly prevalent. However, this upsurge in usage raises significant questions about the transparency and interpretability of the predictions made by these LLM Models. Hence, this paper explores the imperative of ensuring transparency in the application of ChatGPT for public sentiment analysis. To tackle these challenges, we propose using a lexicon-based model as a surrogate to approximate both global and local predictions. Through case studies, we demonstrate how transparency mechanisms, bolstered by the lexicon-based model, can be seamlessly integrated into ChatGPT’s deployment for sentiment analysis. Drawing on the results of our study, we further discuss the implications for future research involving the utilization of LLMs in governmental functions, policymaking, and public engagement.