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

Short summary

HG-RAG is a framework that improves RAG by traversing hierarchical knowledge graphs instead of flat document stores, resolving entity anchors and expanding context through parent, sibling, and child nodes. Evaluated across three world scales and four query types, it consistently outperforms dense retrieval baselines on hierarchical, relational, and multi-hop reasoning while reducing hallucination. The approach is particularly effective when queries require structured or relational reasoning that flat retrieval cannot capture.

• •HG-RAG traverses hierarchical knowledge graphs for structured context retrieval
• •Outperforms flat dense retrieval on hierarchical, relational, and multi-hop queries
• •Reduces hallucination and maintains locality coherence across 18-800 node graphs

Generated with AI, which can make mistakes.