GROUNDEDKG-RAG: Grounded Knowledge Graph Index for Long-document Question Answering

Researchers introduced GroundedKG-RAG, a new retrieval-augmented generation system that creates knowledge graphs directly grounded in source documents to improve long-document question answering. The system reduces resource consumption and hallucinations while maintaining accuracy comparable to state-of-the-art models at lower cost.

• →GroundedKG-RAG addresses key limitations of current RAG systems including high resource consumption and hallucinations from limited grounding.
• →The system constructs knowledge graphs with nodes representing entities and actions, and edges representing temporal or semantic relations.
• →Knowledge graphs are built using semantic role labeling and abstract meaning representation parses from source documents.
• →Performance matches state-of-the-art proprietary long-context models while being more cost-effective.
• →The approach provides interpretable and human-readable results that facilitate auditing and error analysis.