文献类型：期刊文献

英文题名：Reason-Align-Respond: Aligning LLM Reasoning With Knowledge Graphs for KGQA

作者：Shen, Xiangqing[1];Wang, Fanfan[2];Yang, Zinong[2];Wang, Bing[3];Du, Wenli[3];Zong, Chengqing[4];Xia, Rui[1]

机构：[1]Nanjing Univ, Sch Intelligence Sci & Technol, Suzhou 215163, Peoples R China;[2]Nanjing Univ Sci & Technol, Sch Comp Sci & Engn, Nanjing 210094, Peoples R China;[3]East China Univ Sci & Technol, Sch Informat Sci & Engn, Shanghai 200237, Peoples R China;[4]Chinese Acad Sci, Inst Automat, Beijing 100190, Peoples R China

年份：2026

卷号：48

期号：7

起止页码：7340

外文期刊名：IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE

收录：;EI(收录号：20260820137669);WOS:【SCI-EXPANDED(收录号:WOS:001786026100013)】；

基金：This work was supported by the Natural Science Foundation of China under Grant 62476134. Recommended for acceptance by G. Kim.

语种：英文

外文关键词：Cognition; Knowledge graphs; Reliability; Question answering (information retrieval); Semantics; Planning; Natural languages; Large language models; Grounding; Computational efficiency; Knowledge graph question answering; knowledge graph reasoning; large language model; latent variable model; expectation maximization

摘要：Large language models (LLMs) have demonstrated remarkable capabilities in complex reasoning tasks, yet they often suffer from hallucinations and lack reliable factual grounding. Meanwhile, knowledge graphs (KGs) provide structured factual knowledge, but lack the flexible reasoning abilities of LLMs. In this paper, we present Reason-Align-Respond (RAR), a novel framework that systematically integrates LLM reasoning with knowledge graphs for knowledge graph question answering (KGQA). Our approach consists of three key components: a Reasoner that generates human-like natural language reasoning chains, an Aligner that maps these chains to valid KG paths, and a Responser that synthesizes the final answer. We formulate this process as a latent variable mixture model and optimize it using the Expectation-Maximization algorithm, which iteratively refines the reasoning chains and knowledge paths. Extensive experiments on multiple benchmarks demonstrate the effectiveness of RAR, achieving state-of-the-art performance with Hit scores of 93.3% and 91.0% on WebQSP and CWQ respectively. Human evaluation confirms that RAR generates high-quality, interpretable reasoning chains well-aligned with KG paths while maintaining computational efficiency during inference.