arXiv / 2026-11-13

AutoOntic: Automatic Ontology Induction from Text for Knowledge Graph Construction

arXiv:2611.15467v1 Announce Type: new Abstract: Graph-based retrieval-augmented generation depends on the quality of the knowledge graph (KG) it retrieves over, yet schema-free LLM-based extraction pipelines typically produce flat graphs: deduplicated entities and relations with no type hierarchy and no domain/range constraints. We present AutoOntic, a pipeline that automatically constructs an ontology from text by canonicalizing open-vocabulary entity types and relations, inducing a hierarchy o