Ger: Generation, Evaluation and Reflection Enhanced LLM for Knowledge Graph Question Answering
Keywords: KGQA, Reasoning in Large Language Models, Reflection
Abstract: Knowledge Graph Question Answering (KGQA) involves answering natural language questions based on information provided by knowledge graphs. Large language models (LLMs), utilizing their exceptional natural language understanding capabilities and factual knowledge from knowledge graphs, have made some progress in KGQA reasoning. However, existing methods overlook the amplification of hallucinations in large language models caused by irrelevant information within vast knowledge graphs. This oversight leads to answers containing seemingly correct but unrelated responses, decreasing reliability. In this paper, we propose $\textbf{\textit{Generation-Evaluation-Reflection}} $ (Ger), an LLM-enhanced reflective reasoning framework for KGQA. The Ger mechanism introduces evaluation and reflection steps during the reasoning process, allowing LLMs to leverage the factual information in KGs better and utilize their logical reasoning strengths. This process reduces errors and hallucinations while improving reasoning accuracy. Extensive experiments on multiple KGQA benchmark datasets demonstrate that Ger enhances reasoning performance, providing more reliable and interpretable results, and achieves new state-of-the-art.
Primary Area: foundation or frontier models, including LLMs
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: 711
Loading