Datensatz

Zusammenfassung

Tsunami is a typical secondary disaster that can cause unpredictable consequences. After a large earthquake strikes, multiple disaster chains occur when one event triggers another, resulting in a wider and more severe impact than the initial strike. Although some studies have been conducted to analyze cascading effects, many potential impacts of cascading disasters remain uninvestigated from a social sensing perspective. In this study, we propose a method for identifying cascading disasters using natural language processing to extract causal relations comprehensively and objectively between events based on tweets and other auxiliary data, such as remote sensing and newspaper data. The applicability of this method is illustrated by employing data extracted from the Twitter official API on the 2011 Great East Japan earthquake and tsunami. Our methodology can be undertaken in four steps: (i) use relevant keywords to acquire data from Twitter API for the analysis period. (ii) clean and visualize the data, then, extract disaster causal knowledge via pattern matching. (iii) compare semantic and syntactic features and classify the casual and noncausal types using machine learning. (iv) create a disaster keywords database and get tsunami cascading event automatically by calculating cosine similarity between each sentence pair.Our method can effectively and automatically extract the broadest possible range of events during the early stages of the disaster by developing a tsunami event’s relationship. Furthermore, the feature combinations that are most suitable for tweets can also be compared in this paper.