Flood and Weather Monitoring Using Real-time Twitter Data Streams

Social media data is a widely used source to making inference within public crisis periods and events in disaster times. Specifically, since Twitter provides large-scale data publicly in real-time, it is one of the most extensive resources with location information. This abstract provides an overview of a real-time Twitter analysis system to support flood preparedness and response using a comprehensive information-centric flood ontology and natural language processing. Within the scope of this project, we deal with acquisition and processing of real-time Twitter data streams. System fetches the tweets with specified keywords and classifies them as related to flooding or heavy weather conditions. The system uses machine learning algorithms to discover patterns using the correlation between tweets and Iowa Flood Information System's (IFIS) extensive resources. The system uses these patterns to forecast the formation and progress of a potential future flood event. While fetching tweets, predefined hashtags are used for filtering and enhancing the relevancy for selected tweets. With this project, tweets can also be used as an alternative data source where other data sources are not sufficient for specific tasks. During the disasters, the photos that people upload alongside their tweets can be collected and placed to appropriate locations on a mapping system. This allows decision making authorities and communities to see the most recent outlook of the disaster interactively. In case of an emergency, concentration of tweets can help the authorities to determine a strategy on how to reach people most efficiently while providing them the supplies they need. Thanks to the extendable nature of the flood ontology and framework, results from this project will be a guide for other natural disasters, and will be shared with the community.