GEO-SEMANTICS ANALYSIS OF ENVIRONMENTAL DISASTERS IN NIGERIA USING NATIONAL PRINT MEDIA DATA FOR DISASTER MANAGEMENT (Papers Track)
Benedict Ajanaku (Data Science Nigeria); Rashidat Sikiru (Data Science Nigeria); Anthony Soronnadi (Data Science Nigeria); Ife Adebara (Data Science Nigeria); Olubayo Adekanmbi (Data Science Nigeria AI (DSNai))
Abstract
In recent years, Nigeria has experienced various natural and environmental disasters, including floods, food insecurity, fire outbreaks, oil spills, and banditry. These events have caused extensive damage, disrupted lives and properties, and displaced many humans, with emergency response efforts often hindered by the lack of accurate and up-to-date disaster location information. In addressing this gap, we investigated the application of geo-semantics analysis on environmental disasters in Nigeria using media data to map locations and to improve emergency response planning. We developed a disaster location-based (DLB) NER model by fine-tuning three Named Entity Recognition (NER) models—spaCy, BERT, and DistilBERT—with an extensive compiled dataset of disaster-related Nigerian news articles. Each model was evaluated using three metrics, with BERT achieving the highest performance: precision of 0.99331, recall of 0.99349, and f1-score of 0.99297, followed by DistilBERT with precision of 0.99236, recall of 0.99297, and f1-score of 0.99240, and spaCy with precision of 0.95, recall of 0.77, and f1-score of 0.85. The model was used to recognize toponyms and extract location details. Using Nominatim, we resolved the toponyms into coordinates and visualized disaster hotspots. These results show that the fine-tuned NER models can be used in providing precise, real-time mapping, and improving situational awareness for focused interventions. Our approach provides a transformative framework for incorporating print media data into emergency response strategies and informing humanitarian assistance efforts more effectively.