CoTKR: Chain-of-Thought Enhanced Knowledge Rewriting for Complex Knowledge Graph Question Answering

TL;DR

This paper introduces CoTKR, a chain-of-thought based knowledge rewriting method for complex KGQA, which improves the relevance and quality of knowledge used by LLMs, leading to better question answering performance.

Contribution

The paper proposes CoTKR, a novel reasoning trace-based knowledge rewriting method, and PAQAF, a training strategy to align knowledge rewriting with QA model preferences, enhancing KGQA accuracy.

Findings

CoTKR outperforms previous rewriting methods in KGQA tasks.

The approach significantly improves LLMs' accuracy on KGQA benchmarks.

Experimental results validate the effectiveness of the proposed methods.

Abstract

Recent studies have explored the use of Large Language Models (LLMs) with Retrieval Augmented Generation (RAG) for Knowledge Graph Question Answering (KGQA). They typically require rewriting retrieved subgraphs into natural language formats comprehensible to LLMs. However, when tackling complex questions, the knowledge rewritten by existing methods may include irrelevant information, omit crucial details, or fail to align with the question's semantics. To address them, we propose a novel rewriting method CoTKR, Chain-of-Thought Enhanced Knowledge Rewriting, for generating reasoning traces and corresponding knowledge in an interleaved manner, thereby mitigating the limitations of single-step knowledge rewriting. Additionally, to bridge the preference gap between the knowledge rewriter and the question answering (QA) model, we propose a training strategy PAQAF, Preference Alignment from Question Answering Feedback, for leveraging feedback from the QA model to further optimize the knowledge rewriter. We conduct experiments using various LLMs across several KGQA benchmarks. Experimental results demonstrate that, compared with previous knowledge rewriting methods, CoTKR generates the most beneficial knowledge representation for QA models, which significantly improves the performance of LLMs in KGQA.