Fact Extraction via Canonical Identifiers: Validating a Multi-Registry Pipeline for Dutch Open Data

Key Points

• The aim is to validate a multi-registry pipeline for extracting facts using various Dutch canonical identifiers.
• Used seven types of Dutch canonical identifiers (BIG, ECLI, KvK, SKJ, QID, BRIN, AGB) for extraction.
• Compared precision and recall for different identifier types.
• Evaluated regex methods against spaCy Named Entity Recognition (NER).
• Demonstrated cross-registry verification for accountability applications.
• Achieved varying precision and recall metrics across different identifier types.
• Showed that spaCy NER outperformed regex for certain identifiers.
• Confirmed the efficacy of cross-registry verification in enhancing accountability.