A Multi-Dimensional Analysis of User Classification, Sentiment, and Network Influence in Tornado-Focused Social Media Communication

Abstract

During major disaster events, social media serves as a critical platform for user interactions, real-time updates and resource coordination. However, the rapid spread of disaster risk information complicates crisis communication. This study systematically categorized social media users based on their engagement behaviors. The study classifies users into Bot and Non-Bot (automated and human-operated accounts, respectively), with Non-Bot users further subcategorized as Individuals, Public and Private Agencies, Media and others. These categories were analyzed across sentiment, discussion topics, temporal activity patterns, and network interactions. Using advanced natural language processing and network science methods, this study examined geotagged tornado-related posts on X from Oklahoma between 2020 and 2022 and compared trends across years. Results show that social media communication is predominantly citizen-driven, with individuals occupying the most central network positions, media acting as secondary hubs, and bots exerting limited influence. Sentiment patterns further reveal differentiated stakeholder roles. These findings offer actionable insights for emergency managers to improve communication accuracy and coordination.