Abstract Book of the 7th International Conference on New Trends in Management, Business and Economics
Year: 2025
[PDF]
Disaster Management Using Social Media Images: A Case Study for the 2023 Kahramanmaraş Earthquakes in Türkiye
Sema Degirmen-Bektas, Fatih Cavdur, Tülin Inkaya
ABSTRACT:
In disaster management, artificial intelligence and data analytics are important tools for governments and non-governmental organizations to formulate effective response strategies. During and immediately after a disaster, affected people frequently use social media to send real-time aid requests including images of damaged building, infrastructure or injured people. The critical role of social media was particularly observed during the February 6, 2023 Kahramanmaras earthquakes in Türkiye. Motivated by these, this study focuses on the analysis of social media imagery from the Kahramanmaras earthquakes collected from X (formerly Twitter). Considering the importance of rapid image analysis for immediate response during and immediately after a disaster, zero-shot learning was applied as a solution approach. Zero-shot learning is a machine learning technique in which a model can make classification without direct labelled training samples for the target classes. In this context, the CLIP model was used to classify disaster-related images into three categories, i.e. people, building, and textual content. The performances of different CLIP architectures were compared, and the ViT-B/32 model achieved the best classification result with 90% accuracy, 87% precision and 85% recall. This study is a preliminary work for transforming social media content into actionable insights for disaster management.
Keywords: CLIP, deep learning, disaster management, image classification, zero-shot learning