Data Mining and Sentiment Analysis of Real-Time Twitter Messages for Monitoring and Predicting Events

Real-time geolocation data can be used to map natural and social hazards, facilitating predictions and preventive actions. Currently, organizations such as the U.S Geological Survey, the U.S. Center for Disease Control, Weather.com and Google Trends attempt to provide information, but at a high cost and with limitations. This research represents an effort to overcome some of these challenges by offering an inexpensive alternative based on analyzed real-time data with precise geolocation, and including a dashboard informed by human factors analysis. Twitter is a social networking service that allows access to messages and geolocations created by members of the public. For this research, data was collected from Twitter and filtered through data mining techniques based on keywords and phrases. To eliminate tweets with a context other than the focus of this research, sentiment analysis was performed using machine learning algorithms. Visual representations of the results were created including maps of precipitation and earthquakes, and a dashboard showing flu spread. Human factors analysis techniques, mouse and eye trackers, and a small survey-based study were used to verify that users could make accurate interpretations and conclusions.