Bayesian Dynamic Network Modeling for Social Media Political Talk

Abstract

Streaming social media network data have been used in recent studies on political behavior and institutions. Modeling time dynamics in such data helps political scientists produce robust results and efficiently manage their data collection process. However, existing political science methods are yet to provide researchers with the tools to analyze and monitor streaming social media network data. In this thesis, I introduce Bayesian dynamic network modeling for political science research. An extension of the recent development of dynamic modeling techniques, the method enables political scientists to track trends and detect anomalies in streaming social media network data. I illustrate the method with an application to an original dataset of political discourse from a Chinese social networking site. The model detects citizens' behavioral responses to political and non-political events. It also suggests the Chinese government censors and fabricates online discourse during politically sensitive periods.