Machine learning prediction models for the popularization and dissemination of medical science popularization videos

Abstract

Objective To summarize the current shooting trends of this type of video, discuss the effect of non-medical factors on the spread of videos, and develop prediction models using machine learning (ML) algorithms. Methods We searched and filtered medical science popularization videos on TikTok, then labeled non-medical features as variables and record the number of “Thumb-Up”, “Comment”, “Share” and “Collection” as outcome indicators. A total of 286 samples and 34 variables were included in the construction of the ML model, and 13 algorithms were employed with the area under the curve (AUC) for performance assessment and a ten-fold cross-validation for accuracy testing. Results In the quantitative analysis of the 4 outcome indicators, we identified significant disparities among different videos. Subsequently, five best-performing models were ultimately confirmed to predict the reasons for differences: “Thumb-Up” RF Model (AUC = 0.7331), “Collection” RF Model (AUC = 0.7439), “Share” RF Model (AUC = 0.7077), “Comment” RF Model (AUC = 0.7960), “Comment” BNB Model (AUC = 0.7844). By ML models, the video duration, title and description length, shooting location emerged and body language as the most five crucial parameters across all five models. Conclusion ML models demonstrated superior performance in predicting the influence of non-medical factors on the spread of medical science popularization videos. The weight of these variables will provide valuable guidance for video preparation. This study contributes to the dissemination and acceptance of medical science popularization videos by the public, thereby promoting health education and enhancing public awareness and competence in healthcare.