Paragraphs were selected via stratified sampling (random seed 42), capped at 10–25 paragraphs per biography depending on corpus size. Annotations follow the ObliquER pipeline schema with character-level offsets, entity IDs, Wikidata linking, and OOKB classification.

LLM-based entity recognition and linking pipeline designed for implicit and long-tail entities. Features formal task definitions grounded in the Viewsari ontology, prompt engineering for zero-shot and few-shot extraction, dynamic chunking, entity linking with OOKB tagging, and coreference clustering.

The populated KG covers the full Lives corpus: person entities from the index of names, co-occurrence instances with PMI and Dice scores, Web Annotation layer for paragraph-level mention anchoring, provenance activities for each extraction run, and FRBR-level bibliographic metadata.