Generative Ontology Induction: Domain-Agnostic Schema Discovery from Document Corpora Using Large Language Models

Short summary

Generative Ontology Induction (GOI) is a domain-agnostic framework that uses LLMs to extract structured ontologies—entities, properties, relationships, and constraints—from document corpora and export them as typed graphs in YAML/JSON. Validation across four contrasting ontologies (invoices, job descriptions, clinical pain-management records, and professional services contracts) shows 95-100% structural backbone coverage. The authors introduce the Node Coverage Score metric and demonstrate that GOI significantly outperforms generic templates on complex domains.

• •Framework induces structured ontologies from document corpora using LLMs, exporting typed graphs in YAML/JSON
• •Validated on four domains including clinical records and contracts with 95-100% structural coverage
• •Introduces Node Coverage Score as a novel evaluation metric for ontology completeness

Generated with AI, which can make mistakes.