Monitoring a Community's Emotional Landscape: Harnessing Social Media Data to Enhance Telemedicine Services for Mental Health During Disaster

ABSTRACT Disasters severely impact community mental health, as exemplified by the COVID‐19 pandemic. While the pandemic spurred social media and telemedicine adoption, a crucial gap persists in integrating them for assistance in community‐as‐a‐patient care. To bridge this, we propose a novel methodology to correlate community emotional changes with external events during a disaster. For this, we curated two datasets for an urban community for the COVID‐19 period: UMBlog (microblogs) and EvLog (real‐world on‐ground events). We fine‐tuned EmoRegressor which predicts emotion intensities for anger , fear , joy , and sadness from text (Pearson Correlation coefficient and Mean Square Error ). Applying EmoRegressor on UMBlog generated collective emotion time series. These emotion time series, alongside with event time series for EvLog are analyzed to reveal synchronized patterns and significant emotion transitions using change point detection and time coherence analysis. Subsequently, time‐proximity analysis and event attribution yield Attribution Score () values for events triggering emotional changes. To validate robustness, we conducted internal consistency check involving Large Language Models (LLMs) assessment and independent news validation. Further, we provide ‐informed recommendations for telemedicine to plan Psychological First Aid (PFA) targeting community‐as‐a‐patient care. Our analysis underscores the synergy of social media and computational methods with telemedicine in fostering a responsive healthcare system at community scale.