fig4

Figure 4. Dual-pathway LLM-driven ontology engineering. In the top-down route, experts use LLMs as drafting assistants under competency questions and reasoner validation. In the bottom-up route, LLMs induce candidate schemas and populate knowledge graphs from literature. The central reliability principle is mutual constraint: LLMs propose candidates, while ontologies, source evidence, formal validation, and expert review determine admission. LLM: Large language model; RAG: retrieval-augmented generation.