Adolescent engagement and sentiment toward reproductive-health videos on Chinese social media: a cross-sectional content analysis

The study characterises the engagement structure of Bilibili reproductive-health videos: views correlate strongly with likes (r = 0.7), collections (r = 0.74), and shares (r = 0.74), and collections and shares are almost linearly related (r = 0.9). Earlier discussion of Chinese adolescent reproductive-health content online largely noted platform presence; this work provides a quantified correlation structure across 743 videos and 486.8 million views. Based on metadata for 743 videos using descriptive and Pearson-correlation analyses; the sample is large but the design is cross-sectional and correlational.

Less frequently covered topics attract disproportionately high like, share, and comment rates, for example sterilisation surgery at 3.0% of videos and fertility-awareness-based methods at 2.3%. By placing topic coverage alongside interaction intensity, it points to a mismatch between information supply and adolescent interaction demand. Descriptive comparison of topic distribution and engagement metrics; no causal test is reported.

Uploader expertise is limited: only 17% had confirmed medical training, and none were identified for videos on novel contraceptive methods or spontaneous abortion. Manual verification of uploaders' medical backgrounds turns the credibility question from speculation into a checkable count. Manual verification of uploader qualifications, coded across the 743 videos.

Sentiment varies by topic: induced abortion (52.2% negative) and novel contraceptives (52.1% negative) drew mainly negative reactions, while content on coping with spontaneous abortion had the highest proportion of positive sentiment (24.2%); negative sentiment correlated moderately with larger comment volume (r = 0.32) and inversely with likes (r = -0.38) and virtual-coin donations (r = -0.51). Applying a large language model to roughly 760,000 comments and danmaku for three-class sentiment links emotional response to engagement metrics, complementing studies that examine content alone. Sentiment classification covers all comments and danmaku, a large sample; labels are model-generated by Qwen2.5-32B and the correlations are observational.

The study applies to reproductive-health videos retrieved with 18 keywords on Bilibili between September 2016 and June 2024, together with their comments and danmaku, primarily for a user base mostly aged 24 or younger; its engagement correlation structure and sentiment distribution can inform content review, health-communication design, and uploader credential labelling for health educators, platform operators, and clinical communication teams working in similar video communities.

Sentiment classification was performed automatically by Qwen2.5-32B, so the degree of agreement between its labels and human judgement, and contextual differences in emotional expression across topics, remain questions a careful reader would watch; in addition, this reading is at summary scope, so breakdowns in figures and supplementary materials are not included, and the specific numerical relationships between topic coverage and interaction rates await confirmation in the original tables.