AI summaryⓘ
The authors address the problem of finding multiple connected facts (triplets) in knowledge graphs to answer complex questions, a process called multi-hop retrieval. They note that existing methods often treat each fact separately and lack good examples that show how queries relate to each fact, which leads to missing important but loosely connected information. To improve this, the authors created KAMR, a method that identifies key facts directly related to the query and then finds other connected facts that support the answer. They also generate new training data using large language models to better teach the system how queries align with facts. Their approach consistently improves retrieval and question answering on several benchmarks.
multi-hop retrievalknowledge graphretrieval-augmented generationtripletsquery alignmentcontrastive learninglarge language modelsquestion answeringanchor tripletsknowledge-aligned retriever
Authors
Xiaochen Wang, Yuan Zhong, Haoyu Wang, Ting Wang, Fenglong Ma
Abstract
Graph-based retrieval-augmented generation increasingly relies on multi-hop retrieval, where answering a query requires composing multiple connected knowledge-graph triplets. However, existing retrievers often rank triplets independently via global semantic matching. Moreover, many multi-hop benchmarks provide only final answers, which limits supervision for query--triplet alignment and causes structurally necessary but weakly aligned facts to be missed. To address these issues, we propose a knowledge-aligned multi-hop retriever, KAMR, which distinguishes anchor triplets that are strongly constrained by the query from connected triplets that are weakly aligned yet structurally linked to the anchors. To mitigate the lack of query-triplet alignment supervision, we build a partial alignment dataset by masking triplet elements and prompting an LLM to generate corresponding queries, and optimize two contrastive objectives for pair-level and element-level matching. At inference time, KAMR retrieves anchors globally and then expands locally to collect connected evidence. Across four benchmarks, three LLM backbones, and fourteen baselines, KAMR consistently improves multi-hop retrieval and downstream question answering performance.