Deep Learning Models for Predicting Globally Disruptive Events
Principal Investigator: Anna Buczak, Johns Hopkins University APL
Co-Investigators: Mark Dredze, Johns Hopkins University
Years of Award: 2019-2021
Managing Service Agency: Office of Naval Research
Disruptive events -- such as unexpected changes in government and conflicts between governments and local populations -- present significant challenges to United States (US) foreign policy and military posture. Response strategies informed by evidence-based research are increasingly needed to protect US interests. The predictive capability that we are developing developed requires methods for data fusion and prediction (they reside in the prediction engine). It also requires appropriate analytics to characterize historical & social context (H/SC), event similarity (ES), and social discourse (SD). H/SC derives features from historical and socioeconomic data, for example data coming from the World Bank. SD deals with near-real-time data, such as Twitter. ES is a novel measure of correlation between historical and current events. We will be investigating whether social media monitoring that produces realtime information on social discourse is helpful for the prediction of globally disruptive events. We will transform social discourse information into reliable indicators of disruptive events with historical and social context via a deep learning framework. The result of this project will yield a transformative ability to anticipate globally disruptive events. We plan to concentrate in years 1 and 2 on riots and protests that resulted in fatalities.
Anna L. Buczak, Benjamin Baugher, Christine Martin, Meg Keiley-Listermann, James Howard II, Nathan Parrish, Anton Stalick, Daniel Berman, Mark Dredze, “Crystal Cube: Forecasting of Disruptive Events“, submitted to 2020 International Conference on Machine Learning and Data Mining, New York, USA.