Democratizing social media for health research: LLM-powered data analytics platform for NCDs.

Résumé

Despite over 41 million annual deaths from Non-Communicable Diseases (NCDs) globally, predominantly in low and middle-income countries, public access to relevant information from social media is hindered by restrictive licensing of existing social listening tools. This study introduces NCDs Listener, an open-source tool designed to simplify the extraction, summarization, and visualization of NCD-related knowledge from social media comments (Facebook and Reddit posts) in both English and Thai. The tool utilizes keyword matching and the BERT model for knowledge extraction, followed by descriptive statistical analysis. A generative AI model, specifically Google Gemini 2.0 Flash as per the saved information, summarizes this extracted knowledge into human-readable sentences, focusing on medical and healthcare insights. Preliminary results indicate that NCDs Listener improves dashboard comprehension for both general users and data scientists, with the general users showing higher comprehension. Furthermore, both user groups preferred medically focused generative AI summaries over general summaries (p-value