Towards real-time extraction of cascading effect and spatiotemporal analysis using social media data

Compared with isolated hazard events, cascading disasters have the potential to cause greater harm and more dire outcomes. Improving rapid disaster response requires an understanding of how these crises develop. While some previous studies have examined methods for detecting and evaluating cascading disaster effects, there is currently no completely functional system for identifying and analyzing cascading disaster effects in real-time. To solve this issue, this study proposes a novel approach to construct a systematic framework for automatically extracting cascading disaster effects from social media data, which effectively updated the development of effects with hourly updates. Furthermore, geo-locations extracted from text content were combined with cascading effect outcomes to discuss the findings from both temporal and spatial viewpoints. The 2011 Great East Japan Earthquake was selected to test this approach. Our method effectively identified 436 cascading effects, proving its ability to recognize cascading events when tested with a 500-sentence validation set. The geographic locations extracted from the text provide an accurate representation of the affected areas and effectively indicate which topic of effect occurs in specific regions. By leveraging real-time information from social media, this method generates cascading effects without relying on experts’ prior knowledge or experience, making disaster detection more efficient and timelier. There is great potential to apply this novel approach in the domain of cascading disasters to acquire more detailed information with minimal delay when a hazard occurs.