RAG and Vocabulary Filtering for Ontology Learning via Qwen2.5-14B
August 28, 2026
An offline retrieval-augmented pipeline uses Qwen2.5-14B-Instruct and all-MiniLM-L6-v2 to extract ontologies from text. To prevent hallucinations, the system implements deterministic vocabulary-constrained filtering to ensure generated triples match a target term or type vocabulary.
HOW THIS AFFECTS YOU
●
builderThis provides a blueprint for using small-scale RAG to automate domain-specific ontology extension.
●
researcherYou can leverage vocabulary-constrained filtering to reduce hallucination in structured knowledge extraction.