Harnessing the Tweet Stream: Deep Learning for Natural Disaster Detection

Social media has been the go-to source of news for the modern masses in recent years. The flow of information is unmatched on these sites, where big multinationals maintain servers to handle the user information. Twitter has proved a great tool for its use in the spatial and temporal modeling of events, which can be very efficient in gaining automated details on Natural Disasters. Machine learning techniques for Natural Language Processing (NLP) and Twitter can leverage all this information to create a natural disaster identifier. This research paper primarily focuses on using public tweets to assess the occurrence and impact of natural disasters. Initially, we extracted and filtered tweets from public Twitter accounts for disaster-type content using various disaster keywords. Next, we utilized the Bidirectional Encoder Representation from Transformers (BERT) model to classify the tweets, verifying whether they pertained to a disaster. The results reveal the efficacy of this approach in rapidly assessing the occurrence and impact of natural disasters.