Enhancement Large Language Models Domain Through Ontology-Based Retrieval-Augmented Generation

Large Language Models (LLMs) show strong performance in natural language tasks but are prone to hallucinations, limiting reliability in knowledge-intensive fields such as cultural heritage. This paper presents an Ontology-Based Retrieval-Augmented Generation (OB-RAG) framework that embeds subject–predicate–object triples from domain ontologies into a vector space, retrieving relevant knowledge via semantic search to ground LLM outputs. Unlike traditional RAG using unstructured text, the framework integrates manually and semiautomatically generated ontologies for explicit contextual grounding. A cultural heritage case study illustrates implementation and evaluation. Performance is assessed with quantitative metrics (Faithfulness and Answer Relevancy) and expert validation. Results show the OB prototype outperforms baseline LLMs, reducing hallucinations and improving factual accuracy and contextual alignment. The study offers both an architectural framework and empirical evidence that ontology-based RAG strengthens trustworthiness and user acceptance of LLMs in specialized domains.