Characterizing and Evaluating Mental Health Misinformation on Social Media: A Qualitative and Deep Learning-Based Study

Social media dramatically facilitates social communications and information sharing on the Internet. However, social media also widely and quickly spread mental health misinformation and then amplifies the hazards of these misinformation for public health. Therefore, eliminating mental health misinformation on social media is indispensable and urgent. Unfortunately, information on social media is characterized by diversity, real-time, and wide dissemination. It is impossible to manually evaluate and eliminate massive mental health misinformation on social media. To address this challenge, the present study used qualitative and deep learning-based methods to automatically characterize and evaluate mental health misinformation on social media. Specifically, the present study incorporated expert opinions and specific characteristics of social media information to conduct a comprehensive credibility assessment rater guideline through qualitative methods. A detailed 21 fine-grained classification system comprising 7 key dimensions was developed, and a high-quality dataset of 814 social media posts related to mental health was collected by web crawler and then manually annotated according to this system.  Furthermore, this study developed a deep learning-based monitoring framework trained by the dataset for characterizing and evaluating mental health information. In sum, the present study developed three fine tuned deep learning models to efficiently and automatically characterize and evaluate mental health misinformation on social media. These models have the potential to improve the reliability and accuracy of mental health information on social media platforms.