Research proposal: Using Large Language Models to Construct Cross-National Bureaucratic-Politico-Business Networks for Studying Rule of Law Backsliding

This project develops and evaluates an AI-assisted workflow for gathering large-scale, historical, individual-level political, bureaucratic and business networks data across several European countries. The research is embedded in the Horizon Europe project NET-ROL, which aims to investigate the role of networks in law-making, executive decision-making, and the judiciary to better understand the mechanisms behind rule of law backsliding. For this specific part of the project, we examine how Bureaucratic-Politico-Business networks influence the weakening of the rule of law and how these networks, in turn, influence public spending outcomes. The empirical challenge is substantial: identifying and coding the career trajectories, professional ties, political affiliations, organisational memberships, and role-specific connections of top civil servants (e.g., state secretaries, director-generals), regulatory-agency leaders, and high-level public administrators over the past two decades. These individuals are central to public spending decisions, the management of various funds, the design of public procurement procedures, and the distribution of intergovernmental grants. Yet information on their networks is dispersed across thousands of CVs, institutional websites, press releases, legal documents, and procurement records. Manual extraction is thus prohibitively time intensive. The proposed project tests whether and how LLM-based extraction can reliably identify CV-career trajectories, personal assets and interests, and key identifying data to help us infer (1) individual-level networks within the bureaucracy, (2) links between bureaucrats, firms, and political officeholders, and (3) identify networks that might correspond to variation in rule-of-law quality or degradation. The resulting networks will be linked to procurement, company registry and court-decision data. This should enable new research that was previously infeasible on this scale. We think our project would be well-suited to join your proposed Joint-Sessions workshop and contribute to and learn from the discussion around optimal approaches, prompting frameworks, data sources, and validation strategies for extracting richer relationship and network information across a broader range of elites across multiple countries. Authors: Clare Fenwick, Mihaly Fazekas, Jaroslaw Kantorowicz, and Ramin Shirali