A multi-platform comparison and machine learning associative analysis study on the information dissemination effect of short videos for cerebral hemorrhage

Cerebral hemorrhage is a critical public health issue marked by high incidence, disability, and mortality. In China, short-video platforms have become the major channel for the public to acquire relevant health knowledge. However, the online cerebral hemorrhage-related health information varies greatly in quality, with complex dissemination mechanisms and a lack of systematic academic evaluation, which greatly weakens the authenticity and practicality of popular science content for the public. This study aimed to compare the content ecosystems of cerebral hemorrhage-related short videos on TikTok and Bilibili, explore the core factors associated with video dissemination effects, and evaluate the classification performance and feature correlation of machine learning models in identifying high-quality health popularization videos. A total of 200 cerebral hemorrhage-related short videos were collected from the 2 platforms, with their metadata, interaction indicators, and content quality scores recorded. Statistical tests and regression analyses were adopted to compare platform differences and variable correlations. Nine machine learning models were built for high-quality video identification. Area under the curve, calibration curves, and ridgeline plots were used to evaluate model performance and visualize the distribution characteristics of core research indicators. The 2 platforms presented obvious differences in content ecological characteristics. Bilibili had longer-duration, institution-produced, and higher-quality videos, while TikTok featured short videos, individual creation, and stronger user interactivity. Light Gradient Boosting Machine showed the optimal analytical performance, and video duration was the core stable feature affecting video quality. Videos created by medical professionals had significant advantages in information credibility, actionability, and understandability. This study identified a notable "quality-dissemination paradox," revealing the essential conflict between platform algorithm logic and public health information quality. Nonmedical creators need to enhance professional capabilities, and platforms should optimize recommendation algorithms to boost the efficient and accurate dissemination of high-quality cerebral hemorrhage health information.