ToG-I: Progressively Instructed Knowledge Graph-based Large Language Model Reasoning

Minghan Zhang, Zhen Yang, Hongsheng Wu, Yongxing Lin, Jie Chen, Zhen Duan, Shu Zhao

Keywords: LLM Reasoning；Knowledge Graph; Instructed；Progressively

TL;DR: We propose progressive instructed reasoning framework, ToG-I.

Abstract: Large language models (LLMs) reasoning based on knowledge graphs (KGs), by integrating structured knowledge from the KGs, provide a significant solution to alleviate the hallucination problem in complex reasoning tasks. Current techniques mainly focus on the retrieval of explicit knowledge from KGs. LLMs directly use the specific facts and relationships retrieved to construct a reasoning chain to answer the questions. However, these methods often overlook the significance of comprehending implicit knowledge when dealing with problems involving logical reasoning or ambiguous intentions. This could potentially lead to deviations in the reasoning path, hindering their applicability in real-world applications. In this paper, we propose a progressive instructed reasoning framework, ToG-I. The framework identifies core elements, discerns latent intentions, and integrates necessary commonsense reasoning by analyzing the problem from multiple perspectives and levels. Based on this, ToG-I transforms these analysis results into specific reasoning instructions, guiding the LLMs to carry out a progressive reasoning process from a global perspective. This not only ensures the accuracy of the reasoning process but also effectively avoids unnecessary consumption of reasoning resources. Extensive experiments on multiple public datasets show that ToG-I achieves state-of-the-art performance in KG reasoning tasks based on information retrieval and demonstrates superiority in knowledge-intensive tasks.

Primary Area: generative models

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 13943