Mapping Political-Elite Networks in Europe with a Multilingual Joint Entity-Relation Extraction Pipeline

2026-07-27 · Source: cs.CL updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Social Sciences & Behavioral Studies · Depth: Expert, extended

Summary

A new modular, open-weight pipeline for multilingual joint entity-relation extraction is presented, designed to construct signed, temporal knowledge graphs from extensive unstructured news corpora. This architecture combines GLiNER-X-Large for span-based named-entity recognition with a robust three-stage linking cascade to map mentions to language-independent Wikidata identifiers. A Qwen3.6-35B-A3B-FP8 mixture-of-experts model then extracts directed, signed relationships using guided decoding, grounded in a comprehensive domain ontology. Evaluation against a 3,491-relation gold standard demonstrated high textual correctness, ranging from 68.2% strict to 93.7% lenient. Case studies in Austria accurately reconstructed the Alliance for the Future of Austria (BZÖ) party's lifecycle, including its 2009 regional fracture and personnel diaspora, showing a 3.87x higher legal-jeopardy scandal density. In Poland, the system uncovered overlapping economic and governance networks of state-enterprise patronage and the structurally balanced, signed conflict network of the PO–PiS duopoly, with a 2024 SOE leadership turnover spike.

Key takeaway

For research scientists building large-scale computational social science applications, this pipeline offers a robust, replicable framework. You should adopt its modular, open-weight design and ontology-constrained decoding to ensure cross-national comparability and replicability. Integrating external knowledge bases like Wikidata is crucial for language-independent entity resolution. Be aware that entity-linking coverage for mid-tier economic elites remains an area for improvement, requiring careful validation of specific findings.

Key insights

Modular, open-weight pipelines can extract complex, signed political networks from multilingual news, enabling large-scale computational social science.

Principles

• Ontology-constrained decoding ensures consistent output types.
• External knowledge bases anchor entities across languages.
• Modular design localizes extraction failure modes.

Method

A four-stage pipeline (chunking, NER, entity linking, relation extraction) uses GLiNER-X-Large, a three-stage Wikidata linker, and Qwen3.6-35B-A3B-FP8 with guided decoding to build signed, temporal knowledge graphs.

In practice

• Use SKOS ontology for structured, comparable outputs.
• Implement multi-stage entity linking for multilingual text.
• Integrate external structured data for graph enrichment.

Topics

• Joint Entity-Relation Extraction
• Multilingual NLP
• Knowledge Graphs
• Political Elite Networks
• Wikidata
• Large Language Models
• Computational Social Science

Best for: AI Scientist, Research Scientist, NLP Engineer

Related on AIssential

• Mapping Political-Elite Networks in Europe with a Multilingual Joint Entity-Relation Extraction Pipeline — An open-weight pipeline extracts signed, temporal political networks from multilingual news, bridging text to structured data.
• Can an LLM Knowledge Graph Keep Two Unrelated Domains Apart? — The "myKG" pipeline effectively prevents LLM hallucination of cross-domain relationships in knowledge graph extraction through a multi-layered approach.
• Harnessing Structural Context for Entity Alignment Foundation Models — Explicitly leveraging structural context significantly enhances entity alignment foundation models' transferability.
• Conceptual Networks for Cross-Linguistic Idiomatic Expressions:A Feature-Based Graph Approach — Idioms cluster by conceptual schema, not language, enabling cross-lingual transfer and improved detection.
• Construction of Historical Knowledge Graphs Based on BERT and Graph Neural Networks — Combining BERT and GNNs effectively extracts entities and relationships from complex historical texts for knowledge graph construction.
• DECODEM: Data Extraction from Corporate Organizational Documents via Enhanced Methods — Large language models can accurately extract legally meaningful variables from complex corporate documents, though performance varies by provision complexity.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CL updates on arXiv.org.