Multimodal Social Media Data Mining For Disaster Management

Keywords: social media, data mining, clip model, natural disaster management
Abstract Type: Paper Abstract

Abstract

With the recent advances in geo-analytics and geo-AI techniques, the potential of social media data in natural disaster management has been widely recognized. Taking Twitter as an example, the user-generated contents as well as the photo images include rich information about the hazard, benefiting the practical activities for disaster resilience. Relevant studies have witnessed the great capabilities of social media data, while a more comprehensive understanding of these data using emerging machine learning technologies needs further exploration. Utilizing the Contrastive Language–Image Pre-training (CLIP) model, this case study provides a comprehensive perspective to integrate the image and text in user-generated Twitter data, resulting in a more effective approach to extracting key information and enhancing the practices in disaster management.