Student: Carrie Winterer
Advisor: Dr. Alethea Barbaro
Thesis Title: Predicting Twitter Time Series Using Generalized Linear Models
Abstract: Across generations, Twitter is a popular means of communication during the age of social media. Examining the Tweets of a community can provide information about what members of that community are discussing, including issues of social justice and civil unrest. We propose examining the time series occurrence of Tweets, instead of parsing through the content of every Tweet for information. The hypothesis proposed in this thesis is that a linear model can be constructed to predict the number of Tweets per hour in a given community. The occurrence of civil unrest within that community can be determined when the observed number of Tweets deviates from the prediction. After finding that normal linear models do not fit our data satisfactorily, we apply generalized linear models. A bootstrap computation method is used to produce the prediction intervals of these generalized linear models. After building a model that fits the data reasonably, we test our hypothesis on Twitter data from the period of civil unrest that occurred in Baltimore, Maryland in 2015.