Short video platforms have become an important channel for doctors to conduct health science popularization. This study aims to explore the factors influencing health knowledge co-construction on short video platforms. Based on arousal theory, we develop an empirical model to examine the effects of video vividness, video complexity, emotional attitudes, and doctor popularity on health knowledge co-construction. We collected data from the largest short video platform in China, comprising a total of 4,009 videos and 1,040,230 comments for empirical analysis. This study combined machine learning, large language models, and econometric models to test the theoretical model. The results show that: (1) video vividness has a significant positive effect on health knowledge co-construction; (2) video complexity has a negative effect on health knowledge co-construction; (3) there is an inverted U-shaped relationship between emotional attitude and health knowledge co-construction; (4) doctor popularity positively moderates this curvilinear relationship. This study contributes to the literature on knowledge co-construction, expands the understanding of health knowledge co-construction on short video platforms, and provides practical guidance for platform managers and doctors.