Friday, March 7, 8:45 – 10:20 a.m., Mather House 100

Dr. Hyein Ko

Postdoctoral Research Associate, University of Illinois Urbana-Champaign

The presentation will consist of two parts.  Each part will introduce AI/ML tools and methods Dr. Ko has developed and describe how she has used them to  understand politics.  The first part will emphasize the former; the second part will emphasize the latter.

8:45-9:10 Short Presentation on Dr. Ko’s Development of AI/ML Tools and Their Applications

9:15-10:20 “Automated Tracking of Elite Interaction Networks in Authoritarian Regimes”

Elite networks are central to political stability and behavior in authoritarian regimes. To better understand and quantitatively model these networks, we require large-scale, time-sensitive data on ties between actors, which are challenging to conceptualize and measure. This paper addresses these challenges by introducing an automated pipeline centered on large language models (LLMs). Our pipeline includes a new tie ontology with extensive human validation, emphasizing interaction ties between individual actors, as well as tie valence, directedness, and type. Ties are extracted from publicly available news articles, enabling the construction of signed, dynamic networks that evolve over short time scales. We demonstrate that LLMs achieve high accuracy in replicating human annotation, providing an efficient solution for analyzing elite interactions. Applied to Russian elite networks from January 2021 to June 2024, our pipeline shows meaningful changes in the number and valence of elite connections before and after the February 2022 invasion of Ukraine.