ConfliBERT: A Pre-trained Language Model for Political Conflict and Violence

Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 5469 - 5482 July 10-15, 2022 ©2022 Association for Computational Linguistics ConfliBERT: A Pre-trained Language Model for Political Conflict and Violence Yibo Hu†, MohammadSaleh Hosseini†, Erick Skorupa Parolin†, Javier Osorio§, Latifur Khan†, Patrick T. Brandt†, Vito J. D’Orazio† †The University of Texas at Dallas, §The University of Arizona {yibo.hu,seyyedmohammadsaleh.hosseini,erick.skorupaparolin, lkhan,pbrandt,dorazio}@utdallas.edu,josorio1@email.arizona.edu Abstract Analyzing conflicts and political violence around the world is a persistent challenge in the political science and policy communities due in large part to the vast volumes of specialized text needed to monitor conflict and violence on a global scale. To help advance research in political science, we introduce ConfliBERT, a domain-specific pre-trained language model for conflict and political violence. We first gather a large domain-specific text corpus for language modeling from various sources. We then build ConfliBERT using two approaches: pre-training from scratch and continual pre- training. To evaluate ConfliBERT, we collect 12 datasets and implement 18 tasks to assess the models’ practical application in conflict re- search. Finally, we evaluate several versions of ConfliBERT in multiple experiments. Re- sults consistently show that ConfliBERT outper- forms BERT when analyzing political violence and conflict. Our code is publicly available.1 1 Introduction The study of political violence is a central concern of conflict scholars and security analysts in the academic and policy communities. For decades, scholars and governments have devoted incalcu- lable resources to monitoring, understanding, and predicting the dynamics of social unrest, political violence, and armed conflict worldwide. Conflict research is a sub-field of political science that ana- lyzes a broad scope of interactions between govern- ment agents, their challengers, and the civilian pop- ulation, including material and verbal conflict and cooperation. Conflict research covers protest, riots, repression, insurgency, civil war, terrorism, human rights, genocide, criminal violence, forced displace- ment, conventional and unconventional warfare, nu- clear deterrence, peacekeeping, diplomatic disputes and cooperation, among others. 1https://github.com/eventdata/ ConfliBERT Traditionally, researchers used manual coding to track conflict processes worldwide (Raleigh et al., 2010). Unfortunately, the high costs and slow pace of domain experts conducting these tasks make it extraordinarily difficult and costly to monitor highly complex and rapidly changing conflicts in an ever-growing volume of information available on a global scale. Furthermore, these efforts tend to focus on quantifying particular types of conflict events between specific kinds of actors (Sundberg and Melander, 2013). Initial efforts to address these challenges mo- tivated political scientists to develop automated systems to classify or extract structured event data from news articles (Bond et al., 2003; Boschee et al., 2016; O’Brien, 2010; Osorio and Reyes, 2017; Schrodt, 2006, 2009; Alliance, 2015; Norris et al., 2017; Lu and Roy, 2017; Ward et al., 2013). These systems capture a broader range of event types, including different conflict and cooperation events, between a larger number of political ac- tors. They can also extract volumes of data that are orders of magnitude greater than manual coding ef- forts. Automated event data …

Content truncated to 500 words