Decoding Linguistic Ambiguity in Times of Emergency based on Twitter Disaster Datasets

As of November 2020, the massively popular social media site Twitter boasts 1.3 billion accounts, 330 million monthly users, and 500 million tweets per day. The sheer volume of data shared on this platform has legitimized Twitter as a real-time information sharing monolith. Millions of people can share information via Twitter making it's power in this space undeniable. In this project, we'd like to use that power for good. Twitter's real-time information sharing prowess has become the platform of interest when attempting to learn information on events in progress. This paper identifies Tweets related to disasters, accidents, or emergencies and determines such statements' sincerity. We introduce a new model that can detect real emergency tweets with a high classification accuracy by using a transfer learning model BERT for emotion classification and behavior modeling approach to detect tweets in Twitter disaster datasets. In doing so, we hope to provide a great service in identifying relevant, real-time information to first responders.