Insurgency Prediction Using Multiple High-Volume Social Media Data Sources

• Activity Number: 563
• Type: Contributed
• Date/Time: Wednesday, August 12, 2015 : 11:35 AM to 12:20 PM
• Sponsor: Social Statistics Section
• Author(s): Gizem Korkmaz* and Shane Reese and Dave Higdon and Sallie Keller and Naren Ramakrishnan
• Companies: Virginia Tech and Brigham Young University and Virginia Tech and Virginia Tech and Virginia Tech
• Keywords: Twitter ; Civil unrest ; Lasso ; Bayesian

Abstract

Civil unrest events (protests, strikes) unfold through complex mechanisms that cannot be fully understood without capturing social, political and economic contexts. Modern cultures use a variety communication tools (e.g., traditional and social media) to coordinate in order to gather a sufficient number of people to raise their voice. To accommodate dynamic features of social media feeds and their impacts on insurgency prediction, we develop a dynamic linear model based on daily keyword combinations. In addition, due to the large number of so-called n-grams, we employ a sparseness encouraging prior distribution for the coefficients governing the dynamic linear model. Included in the predictors are significant sets of keywords extracted from Twitter, news, and blogs. We include volume of requests to Tor, a widely-used anonymity network, economic indicators and two political event databases (GDELT and ICEWS). Insurgency prediction is further complicated by the difficulty in assessing the exact nature of the conflict. Our study is further enhanced by the existence of a ground truth measure of conflicts compiled by an independent group of social scientists and experts on Latin America.