Structure Guided Retrieval-Augmented Generation for Factual Queries

TL;DR

This paper introduces SG-RAG, a novel retrieval-augmented generation method that uses structural information to improve factual accuracy in complex queries, supported by a large new dataset.

Contribution

It formulates the Exact Retrieval Problem (ERP) incorporating structural info into RAG and proposes SG-RAG to address this, outperforming baselines on a new dataset.

Findings

SG-RAG significantly outperforms baselines on ERQA dataset

Constructed and released the ERQA dataset with 120,000 QA pairs

SG-RAG improves accuracy by 20.68 to 50.88 points across metrics

Abstract

Retrieval-Augmented Generation (RAG) has been proposed to mitigate hallucinations in large language models (LLMs), where generated outputs may be factually incorrect. However, existing RAG approaches predominantly rely on vector similarity for retrieval, which is prone to semantic noise and fails to ensure that generated responses fully satisfy the complex conditions specified by factual queries, often leading to incorrect answers. To address this challenge, we introduce a novel research problem, named Exact Retrieval Problem (ERP). To the best of our knowledge, this is the first problem formulation that explicitly incorporates structural information into RAG for factual questions to satisfy all query conditions. For this novel problem, we propose Structure Guided Retrieval-Augmented Generation (SG-RAG), which models the retrieval process as an embedding-based subgraph matching task, and uses the retrieved topological structures to guide the LLM to generate answers that meet all specified query conditions. To facilitate evaluation of ERP, we construct and publicly release Exact Retrieval Question Answering (ERQA), a large-scale dataset comprising 120000 fact-oriented QA pairs, each involving complex conditions, spanning 20 diverse domains. The experimental results demonstrate that SG-RAG significantly outperforms strong baselines on ERQA, delivering absolute improvements from 20.68 to 50.88 points across all evaluation metrics, while maintaining reasonable computational overhead.