Public Trust in ChatGPT: A BERT-Based Framework for Analyzing Social Media Discourse

Exploring the trust relationship between humans and technology is becoming a core issue in the technological development trend and the future of humanity. The research develops an innovative solution to recognizing public trust in ChatGPT through an analysis of semantic features in textual data. We collected 73,118 comments from December 2022 to May 2024 on ChatGPT to recognize public trust and concern. Bidirectional Encoder Representations from Transformers (BERT) was introduced to improve trust classification accuracy. We employed sentiment analysis and Latent Dirichlet Allocation (LDA) model to understand the concerns of each group. The results indicate that the negative public sentiment toward ChatGPT at different levels of trust dominated. Meanwhile, the sentimental gap in the trusted comments was larger than that in the distrusted group. There is a relationship between media exposure, sentiment and trust in the topic Media. Our findings provide valuable implications for the future development of ChatGPT.