Path Pooling: Training-Free Structure Enhancement for Efficient Knowledge Graph Retrieval-Augmented Generation
Hairu Wang, Yuan Feng, Xike Xie, S Kevin Zhou
TL;DR
This work tackles the gap between efficient triple-based KG-RAG and the richer structural cues offered by path-based methods, which often incur high costs. It introduces a training-free path pooling strategy that uses path kernel search over a local subgraph and mean pooling with a position-aware adjustment to smooth triple scores, making structure information readily available to triple-based KG-RAG. The approach is plug-and-play, featuring two refinement mechanisms—Position Reranking and Triple Reselection—to improve generation quality with minimal retrieval overhead. Empirical evaluation on CWQ across multiple LLMs demonstrates consistent improvements in Hit@1 and Macro-F1, validating the practical impact of incorporating graph structure without additional training.
Abstract
Although Large Language Models achieve strong success in many tasks, they still suffer from hallucinations and knowledge deficiencies in real-world applications. Many knowledge graph-based retrieval-augmented generation (KG-RAG) methods enhance the quality and credibility of LLMs by leveraging structure and semantic information in KGs as external knowledge bases. However, these methods struggle to effectively incorporate structure information, either incurring high computational costs or underutilizing available knowledge. Inspired by smoothing operations in graph representation learning, we propose path pooling, a simple, training-free strategy that introduces structure information through a novel path-centric pooling operation. It seamlessly integrates into existing KG-RAG methods in a plug-and-play manner, enabling richer structure information utilization. Extensive experiments demonstrate that incorporating the path pooling into the state-of-the-art KG-RAG method consistently improves performance across various settings while introducing negligible additional cost.