MDAF Framework Uses LLMs to Automate Foreign Policy Analysis Across Five Eyes

A new research paper from Ma et al. presents MDAF (Multi-Dimensional Annotation Framework), a scalable approach to convert unstructured foreign policy texts from government websites into standardized event data. The framework defines a policy event broadly as any statement or action and leverages an LLM workflow to automate three key tasks: identifying relevant policy texts, extracting structured information, and classifying events. This reduces the manual effort traditionally required for comparative foreign policy analysis.

Empirically, the authors apply MDAF to China-related policy documents from the Five Eyes intelligence alliance (US, UK, Canada, Australia, New Zealand). The resulting database reveals significant variation in how each state frames and implements its China policy, and allows researchers to track temporal shifts in these profiles. The framework contributes to computational international relations by providing a reproducible, AI-driven method for large-scale policy analysis. The paper (32 pages, 9 figures, 5 tables) is available on arXiv.