Know3-RAG: A Knowledge-aware RAG Framework with Adaptive Retrieval, Generation, and Filtering

TL;DR

Know³-RAG tackles hallucinations in open-domain QA by injecting structured knowledge from knowledge graphs into three sequential stages: retrieval, generation, and filtering. It introduces a knowledge-aware adaptive retrieval mechanism that uses KG embeddings to compute a reliability score $s_t$ and a dynamic threshold $\theta_t$ to trigger retrieval, a knowledge-enhanced reference generation step that augments queries with KG-derived entities, and a knowledge-driven reference filtering module that screens references using a relevance check and a triple factual check. Across HotpotQA, 2WikiMultiHopQA, and PopQA, Know³-RAG achieves state-of-the-art or near-state-of-the-art performance and significantly reduces hallucinations, with ablations confirming the contribution of each component. These results demonstrate that explicit KG supervision can enhance factual accuracy and interpretability in RAG systems, offering practical impact for reliable, knowledge-grounded QA.

Abstract

Recent advances in large language models (LLMs) have led to impressive progress in natural language generation, yet their tendency to produce hallucinated or unsubstantiated content remains a critical concern. To improve factual reliability, Retrieval-Augmented Generation (RAG) integrates external knowledge during inference. However, existing RAG systems face two major limitations: (1) unreliable adaptive control due to limited external knowledge supervision, and (2) hallucinations caused by inaccurate or irrelevant references. To address these issues, we propose Know3-RAG, a knowledge-aware RAG framework that leverages structured knowledge from knowledge graphs (KGs) to guide three core stages of the RAG process, including retrieval, generation, and filtering. Specifically, we introduce a knowledge-aware adaptive retrieval module that employs KG embedding to assess the confidence of the generated answer and determine retrieval necessity, a knowledge-enhanced reference generation strategy that enriches queries with KG-derived entities to improve generated reference relevance, and a knowledge-driven reference filtering mechanism that ensures semantic alignment and factual accuracy of references. Experiments on multiple open-domain QA benchmarks demonstrate that Know3-RAG consistently outperforms strong baselines, significantly reducing hallucinations and enhancing answer reliability.

Figures (10)

• Figure 1: Comparison of three different RAG frameworks. Our proposed Know³-RAG Framework employs knowledge graphs to facilitate adaptive retrieval and reference processing.
• Figure 2: Model Architecture of Know³-RAG, which contains: (a) Knowledge-aware Adaptive Retrieval, (b) Knowledge-enhanced Reference Generation, (c) Knowledge-driven Reference Filter.
• Figure 3: Reference Used of Various Knowledge Models in Turn 1
• Figure 4: A case study of Know³-RAG, which utilize knowledge graph for open-domain question.
• Figure 5: Performance of Know³-RAG with Varying Maximum Number of Iterations