Data-Mining Social Media for Spatiotemporal Patterns of Negative Opinion
Given that political and natural events cause unrest in social media, the research of this thesis serves to address the following question: Using census demographics and the geographic signature of social media content, can we determine what effect major political and natural disaster events have on public attitudes towards the American government? The thesis will present my initiative to enrich social media metadata by using the geographic content embedded in each message. Section 2 describes the methodology. First a survey of the social media platforms Twitter, Facebook, and Foursquare is provided. Next it covers methods directly related to geographical theme extraction on social media, along with the most common issues that need to be resolved in order to pursue this line of research. This is followed by a section on the Twitter API and Gnip, the exclusive provider of historical and unlimited Twitter activity. Next I discuss Hurricane Sandy and the 2012 Election, the case studies I decided to test. Then I discuss the data preparation and processing, and the means of temporal and spatial analysis. Section 3 presents the results of my tests and Section 4 concludes this thesis and presents future directions of research areas.
Includes Bibliography, Figures, Maps and Appendix