Towards efficient Graph-RAG via structure-aware intermediate representation: Incremental collaborative exploration on knowledge graph

doi:10.1016/j.knosys.2026.115884

Towards efficient Graph-RAG via structure-aware intermediate representation: Incremental collaborative exploration on knowledge graph

2026-03-27

0

PRE

AI

DOI：10.1016/j.knosys.2026.115884

Original

Original request for help

Save

Abstract

En 中文

• We propose PRIME, a structure-aware intermediate representation framework that aligns natural language sub-queries with KG subgraphs in a unified embedding space, effectively bridging the semantic-structural gap and enabling precise, efficient Graph-RAG. • We introduce an incremental collaborative exploration mechanism that dynamically expands KG regions based on query semantics, reducing redundant search by over 99% while preserving multi-hop reasoning completeness. • PRIME achieves 91.6% Hit@1 on WebQSP and 78.3% on CWQ, outperforming state-of-the-art baselines by up to 4.5%, with inference time is more than 1000 times faster than training-based methods, demonstrating superior accuracy, efficiency, and interpretable reasoning paths.

Keywords:

PRIME

Graph-RAG

knowledge graph

intermediate representation

incremental collaborative exploration

Towards efficient Graph-RAG via structure-aware intermediate representation: Incremental collaborative exploration on knowledge graph

Abstract

Journal

Organization