Short-Form Videos and Mental Health: A Knowledge-Guided Neural Topic Model

作者：Jiaheng Xie, Yidong Chai, Ruicheng Liang, Yang Liu, Daniel Zeng · 发表于：Information Systems Research · 年份：2025 · DOI：10.1287/isre.2024.1071 · 被引用次数：3 · 研究领域：Computational and Text Analysis Methods、Media Influence and Health、Mental Health via Writing

Short-form video platforms such as TikTok and Douyin are widely used but have sparked serious concerns about their impact on youth mental health, especially suicidal thoughts. This study introduces a novel knowledge-guided neural topic model that predicts a video’s potential to induce suicidal thoughts in viewers at the time of upload. Unlike existing models, our approach integrates medical knowledge on suicide risk factors with user-generated content to improve prediction accuracy and explainability. Tested on real-world data from two major platforms, the model not only outperforms current machine learning and deep learning benchmarks but also uncovers emerging content themes linked to suicidal thought risk. For practice, this tool can be directly integrated into platforms’ content moderation pipelines, identifying high-risk videos for follow-up human review before harm spreads. For policy, it offers a scalable and ethically informed method to mitigate digital risks to youth mental health, balancing user safety with content creator rights. This work offers a critical step forward in responsible AI and public mental health protection in the era of algorithm-driven media.