Graph-Verified Bootstrapping for Multi-Hop Knowledge Reasoning
Multi-hop knowledge reasoning over knowledge graphs remains fundamentally challenging under weak supervision, as lightweight models lack intermediate signals for relation composition, while large language models suffer from hallucinations. We propose a graph-verified bootstrapping framework that exploits the knowledge...