📑 arXiv Apr 22
Automatic Ontology Construction Using LLMs as an External Layer of Memory, Verification, and Planning for Hybrid Intelligent Systems
Proposes a hybrid architecture where LLMs are augmented with an automatically constructed RDF/OWL ontology as an external memory layer, replacing or supplementing vector-based RAG with a structured knowledge graph. The pipeline performs entity recognition, relation extraction, triple generation, and SHACL/OWL validation from heterogeneous sources, enabling persistent and verifiable reasoning. The key distinction from standard RAG is that retrieved context is semantically structured and constraint-validated rather than embedding-similarity ranked.