Detecting Medical Misinformation on Social Media Using Multimodal Deep Learning.

Area of Science:

• Public Health
• Computer Science
• Social Media Analysis

Background:

• Vaccine-preventable disease outbreaks are increasing, linked to social media misinformation.
• Existing antivaccine message detection systems often ignore visual content, limiting their effectiveness on platforms like Instagram.

Purpose of the Study:

• To develop an automatic, multimodal detector for antivaccine messages on social media.
• To address the limitations of text-only detection systems by incorporating visual analysis.

Main Methods:

• A deep learning network was designed to process both visual and textual data from social media posts.
• A novel semantic- and task-level attention mechanism was developed to focus on critical antivaccine content.
• An ensemble method was employed to enhance prediction accuracy.

Main Results:

• The proposed model achieved over 97% testing accuracy on a dataset of over 30,000 Instagram posts.
• The multimodal approach significantly outperformed existing models in detecting antivaccine messages.
• The system demonstrates a high capacity for identifying daily antivaccine content.

Conclusions:

• Multimodal deep learning offers a powerful solution for detecting antivaccine messages.
• The developed model can effectively combat the spread of health misinformation online.
• This technology can support public health initiatives by identifying and potentially mitigating vaccine hesitancy.