Analysis of post-6th February earthquake health needs with text mining: A study based on social media data

The February 6, 2023 Kahramanmaraş earthquakes caused great losses and required a rapid and accurate assessment of health needs. This study addresses this challenge by employing a computational framework-combining LDA, BERTopic, and transformer-based sentiment analysis (BERTurk) and supervised ML- to systematically analyze 88.073 Turkish-language tweets over the two-month post-disaster period. The study aimed to identify, categorize post-earthquake health needs discourse to inform disaster response. The topic modeling successfully categorized health needs, revealing a strong alignment with international humanitarian standards (WHO EMT and SPHERE). A critical finding was the extraordinary salience of Hygiene (β= 0.3326), underscoring the public health risk from damaged Water, Sanitation, and Hygiene (WASH) infrastructure. Furthermore, the temporal analysis confirmed a predictable evolution of needs, transitioning from immediate Emergency Medical Services to sustained Public Health and Shelter concerns. Machine learning evaluation identified Logistic Regression and Linear SVM as the superior models for health content classification, achieving accuracies of 89.07% and 88.92% respectively. While broad categories like Medicine and Supplies achieved an F1-score of 0.93, specialized needs like Maternal and Child Health (F1=0.77) highlighted the challenge of linguistic overlap in multi-class categorization. Sentiment analysis revealed a statistically significant correlation between health-related discourse and distress: health tweets exhibited 7.1 percentage points higher negative sentiment (79.7%) than non-health tweets (p < 0.001). This research validates social media analysis as an effective, near-real-time complement to traditional needs assessment, offering actionable insights for adaptive disaster management strategies.