[ICLR-2024] In this paper, we introduce Reasoning on Graphs (RoG), a novel method that enhances Large Language Models (LLMs) with Knowledge Graphs (KGs) to address their limitations in up-to-date knowledge and reasoning hallucinations, by utilizing KGs for faithful and interpretable reasoning. RoG employs a planning-retrieval-reasoning framework to generate relation paths from KGs, enabling LLMs to perform more accurate reasoning, and has shown state-of-the-art performance on benchmark KG reasoning tasks.
