
Can Large Language Models Accurately Predict An Individual’s Offline Health Behavior Based on a Single Online Comment Without Any Specific Training?
By Prof. Chenwei ZHANG, Assistant Professor, HKU Faculty of Education
The registration link will be available soon. Don’t wanna miss out? Subscribe here
Abstract
Quantifying the effects of information campaigns, particularly the impact of online information on human society, is often challenging. This study explores the potential of using large language models (LLMs) as tools to probe offline health behavior in the context of COVID-19, based on users’ online comments. Two datasets are utilized: one online and one offline. The online dataset consists of users’ comments and the social media posts or videos these comments engage with. The offline dataset is derived from survey data that reveals the corresponding offline health behaviors of the users who made these comments. LLMs are tasked with predicting social media users’ survey responses. Experimental results indicate that while current LLMs struggle to accurately predict specific survey responses about particular health behaviors, they are capable of predicting general health behavior tendencies. This study examines the impacts of four key factors in prompt engineering—task description, contextual information, demonstrations, and template formatting—on the performance of LLMs. Finally, this paper discusses the discrepancy between LLMs’ prior knowledge about human behavior and evolving human values, and proposes strategies to narrow this gap and improve LLMs’ understanding and prediction of human behavior.
About the Speaker

Prof. Chenwei Zhang
Assistant Professor, HKU Faculty of Education
Dr. Chenwei Zhang is an Assistant Professor of Information Science at Faculty of Education, the University of Hong Kong. Her research focuses on the data-driven science of science, where she employs multidisciplinary methods to investigate research practices and the conduct of scientific inquiry. Her work spans individual researchers, scientific teams, and research ecosystems. For individual researchers, she focuses on gender and career stage issues. In examining scientific teams, Dr. Zhang emphasizes the significance of diversity and leadership, as well as contributions and collaborations at a finer-grained level. Her work on research ecosystems delves into research evaluation and the dissemination of scientific findings. Additionally, she utilizes various big data analytics approaches, including machine learning and deep learning, to gain insights into key issues in higher education and educational technology.