Identity and Ontology Tagging in a Production Knowledge Graph

A recent study published on arXiv (2608.10644) outlines an ingestion and ontology-tagging framework for a comprehensive knowledge graph comprising 537,157 entities and 2,198,567 relationships, sourced from 98,795 government documents. The authors contend that decisions regarding identity during the graph's creation can be detrimental, as merging records under a single identity is irreversible once their properties are combined, unlike extraction mistakes that can be rectified later. To mitigate this issue, they introduce a record-identity ladder that assesses sameness through identifier columns, name columns, display names, and type-scoped position instead of relying on name similarity. This ladder facilitates de-duplication in parsed tables, while a broader canonical-name key is used during graph writing, ensuring that records with identical canonical names merge automatically when they are exactly equal. The paper underscores the critical need for meticulous identity management in the construction of knowledge graphs, especially in the context of large-scale document processing.