Using Social Media Data to Predict Conflict Escalation

• In Event: Novel Approaches to Studying Escalation of Political Violence

Abstract

Observational civil war data often fails to capture important on-the-ground variables. Examples include: the opposition’s dedication to the cause, the public’s perceptions of injustice, and levels of civilian fear. These can be important weathervanes for predicting the course of political violence but are difficult to measure using traditional methods. This paper uses social media data to capture these variables by drawing on a sample of 1% of all Twitter posts geolocated in Africa. Topic modeling and sentiment analysis allow us to produce cross-national measures of these intangible variables that account for precise physical and temporal location. By analyzing this data, we examine whether the tone or topics of social media posts can predict civil war escalation at a specific place and/or time. We evaluate the value of social-media data for this purpose by incorporating them in the ViEWS forecasting model of organized political violence and assess the extent to which they improve the model’s ability to forecast the onset, escalation, and de-escalation of violence.

Authors

• Caroline Brandt, Uppsala University

• Håvard Hegre, Uppsala University