Tracking Non-State Armed Actors in Afghanistan Using ML and NLP in Arabic

• In Event: Mapping the Presence and Relationships of Non-State Armed Groups

Abstract

This paper applies new developments in Machine Learning (ML) and Natural Language Processing (NLP) in Arabic language to generate a geo-referenced database on the violent presence of non-state armed actors in Afghanistan. This computational social science application processes a vast collection of news stories written in Arabic to generate a granular database of the territorial presence of the Taliban, Al-Qaeda, ISIS, and Hezb-e-Islami between 2008 and 2018. The unprecedented granularity of the data allows us to disaggregate daily information at the district level for each of these actors. Results of this project come accompanied with an interactive web application for tracking the territorial presence of Afghan non-state armed actors over time and across space. This data is then used to study attacks against government and international forces in Afghanistan.

Author

• Javier Osorio, University of Arizona