arXiv subjects

Rex W. Douglass

Publications and source records attributed to Rex W. Douglass.

4 recordsLinked to original sources

arXiv2024

ICBeLLM: High Quality International Events Data with Open Source Large Language Models on Consumer Hardware

The International Crises Behavior Events (ICBe) ontology provides high coverage over the thoughts, communications, and actions that constitute international relations. A major disadvantage of that level of detail is that it requires large human capital costs to apply it manually to new texts. Whether such an ontolgy is practical for international relations research given limited human and financial resources is a pressing concern. We introduce a working proof of concept showing that ICBe codings can be reliably extracted from new texts using the current generation of open source large language models (LLM) running on consumer grade computer hardware. Our solution requires no finetuning and only limited prompt engineering. We detail our solution and present benchmarks against the original ICBe codings. We conclude by discussing the implications of very high quality event coding of any text being within reach of individual researchers with limited resources.

stat.AP↗

arXiv2024

What is Escalation? Measuring Crisis Dynamics in International Relations with Human and LLM Generated Event Data

When a dangerous international crisis begins, leaders need to know whether their next move is going to resolve the dispute or amplify it out of control. Theories of conflict have mainly served to deepen the confusion, revealing fighting, bargaining, and signaling to be high-dimensional and subtle equilibrium behaviors with deeply contextual consequences. Should a leader communicate resolve through aggressive acts, avoid spirals through accommodation, or focus on ensuring the possibility of a bargain? We offer a data-driven empirical solution to this logjam in the form of a new large-scale analysis of actions taken within 475 crises. We combine two complimentary measurement projects, the human-coded International Crisis Behavior Events (ICBe) dataset and the new machine-coded ICBeLLM. We model directly whether an action tends to shorten or extend the length of a crisis. The result is a directly interpretable measure of the latent escalatory/de-escalatory nature of each action leaders have chosen over the last century.

physics.soc-ph↗

arXiv2022

Introducing the ICBe Dataset: Very High Recall and Precision Event Extraction from Narratives about International Crises

How do international crises unfold? We conceptualize of international relations as a strategic chess game between adversaries and develop a systematic way to measure pieces, moves, and gambits accurately and consistently over a hundred years of history. We introduce a new ontology and dataset of international events called ICBe based on a very high-quality corpus of narratives from the International Crisis Behavior (ICB) Project. We demonstrate that ICBe has higher coverage, recall, and precision than existing state of the art datasets and conduct two detailed case studies of the Cuban Missile Crisis (1962) and Crimea-Donbas Crisis (2014). We further introduce two new event visualizations (event icongraphy and crisis maps), an automated benchmark for measuring event recall using natural language processing (sythnetic narratives), and an ontology reconstruction task for objectively measuring event precision. We make the data, online appendix, replication material, and visualizations of every historical episode available at a companion website www.crisisevents.org and the github repository.

stat.AP↗

arXiv2015

Understanding Civil War Violence through Military Intelligence: Mining Civilian Targeting Records from the Vietnam War

Military intelligence is underutilized in the study of civil war violence. Declassified records are hard to acquire and difficult to explore with the standard econometrics toolbox. I investigate a contemporary government database of civilians targeted during the Vietnam War. The data are detailed, with up to 45 attributes recorded for 73,712 individual civilian suspects. I employ an unsupervised machine learning approach of cleaning, variable selection, dimensionality reduction, and clustering. I find support for a simplifying typology of civilian targeting that distinguishes different kinds of suspects and different kinds targeting methods. The typology is robust, successfully clustering both government actors and rebel departments into groups that mirror their known functions. The exercise highlights methods for dealing with high dimensional found conflict data. It also illustrates how aggregating measures of political violence masks a complex underlying empirical data generating process as well as a complex institutional reporting process.

cs.CY↗