Disaster Information Verification and Validation Application Using Machine Learning

Abstract

Social media platforms have made it possible for people and organizations to disseminate information to their peers and target markets. Even while most information is shared with the best of intentions, some people utilize social media to further their own agendas. They might publish untrue or inaccurate information in their posts. Before, during, and after disasters and emergencies, social media is rife with rumors, misinformation, and misleading information. These false rumors and information could also make individuals anxious. How to stop the spread of this incorrect information is one of the main problems that public safety authorities and organizations face. DIVVA is a system that, using some provided input text data, evaluates and validates disaster-related information. The system has two tracks: a validation track and a verification track. Verification will classify the textual inputinto categories related to disasters or not related to disasters. The Validation track, on the other hand, will use the official handles of government disaster relief organizations like the NDRF (National Disaster Response Force) to determine whether the event mentioned in the text data actually happened or not before classifying the disaster- related data as real or fake. Therefore, if many individuals receive erroneous information about a calamity, we can utilize our approach to determine if the information is true or false. Our results show that the Bidirectional LSTM model performs well for the tweet classification (i.e. whether the tweets are related to disaster or not) task with 84% accuracy.