Mapping Political-Elite Networks in Europe with a Multilingual Joint Entity-Relation Extraction Pipeline: The VALPOP Project

How political elites organise, whether into rent-seeking coalitions that capture public resources or civic networks that sustain democratic governance, is a central question in comparative politics. Addressing the workshop's focus on using AI tools for elite-network research, we present a pipeline developed within the EU-funded VALPOP project that extracts typed, entity-resolved political relationships from multilingual news text using open-weight language models. The pipeline connects extracted entities to Wikidata for cross-language identity resolution and enforces a fixed ontology through constrained decoding, producing cross-national, multiplex, and signed networks. The open and modular architecture separates data, ontology, and models, so that individual components can be replaced while keeping outputs replicable and comparable across countries and time periods. In a pilot study on Polish news (~250,000 articles), the pipeline produces over 1.1 million relations with a correctness rate of 78.5%, evaluated against a 3,491-relation gold standard. The resulting network recovers known features of Polish politics: the governance-economic overlap around state-owned enterprises, the adversarial structure of the PO-PiS cleavage, and a temporal shift in SOE-politician ties consistent with documented post-election patronage turnover. The pipeline is designed to move beyond inner circles of top leaders and reconstruct the broader networks through which elites shape public-good provision in domains such as welfare, energy, and infrastructure, including the mechanisms by which populist actors transform public goods into club goods. Expansion to 12 European countries is underway.