“Short” is not always “scientific”: cross-sectional quality assessment and machine learning–based evaluation of weight management short videos on TikTok and Bilibili

Objective This study aimed to systematically evaluate and compare the content quality and reliability of weight management–related short videos on TikTok and Bilibili, identify key factors influencing video quality, and develop an automated quality prediction tool to support the construction of a healthier digital health communication ecosystem. Methods A cross-sectional study design was adopted. The top 100 weight management–related videos ranked by overall relevance were collected from TikTok and Bilibili, respectively (total n = 200). Two independent researchers evaluated video quality and information reliability using the Global Quality Score (GQS) and the DISCERN instrument. Spearman’s rank correlation and Poisson regression analyses were conducted to explore associations between video characteristics and evaluation scores. An XGBoost-based model was developed to predict high-quality videos, and the SHAP framework was applied to interpret the model’s decision-making mechanism. Results Videos on Bilibili had significantly higher GQS scores than those on TikTok ( p < 0.001). However, DISCERN scores were generally low on both platforms, with no statistically significant difference. On TikTok, most videos were uploaded by Nonprofessional individuals and primarily focused on Personal experience sharing (68%), whereas the distribution of uploader sources on Bilibili was relatively balanced. Videos published by Professional individuals demonstrated significantly higher quality and reliability than those published by Nonprofessional individuals. Regression analyses indicated that video duration and number of likes were positive predictors of both quality and reliability. The XGBoost model achieved good discriminative performance in the test set (AUC = 0.8072). SHAP analysis revealed that when video duration exceeded 12.75 min, its contribution to predicting high-quality videos shifted from negative to positive. Based on these findings, an accessible online evaluation platform for high-quality short videos was developed and deployed. Conclusion Videos published by Professional individuals possess greater academic value and practical significance. However, weight management information on short-video platforms exhibits a mismatch between popularity and quality. The automated evaluation model and online tool proposed in this study provide strong support for the public in identifying reliable scientific information and for regulatory authorities in developing intelligent governance systems.