HG-RAG: Hierarchy-Guided Retrieval-Augmented Generation for Structured Knowledge Graphs

arXiv cs.AI · 6 hours ago

HG-RAG is a retrieval-augmented generation framework that navigates hierarchical knowledge graphs instead of flat document stores to provide structured context to language models. The system was evaluated on three knowledge graph sizes ranging from 18 to 800 nodes across four query types, showing consistent improvements over dense retrieval baselines on hierarchical, relational, and multi-hop reasoning tasks. This approach reduces hallucination and enables language models to perform better on complex reasoning tasks that require understanding relationships across structured knowledge.