Query-Aware Graph Attention for Precise Subgraph Retrieval in Knowledge-Augmented Reasoning

Large language models (LLMs) increasingly rely on external knowledge to mitigate hallucinations, yet retrieving precise multi-hop evidence for knowledge-augmented reasoning remains difficult. Existing Knowledge Graph (KG)-based Retrieval-Augmented Generation (RAG) systems insufficiently model the interaction between query semantics and relation types, resulting in imprecise subgraph retrieval and unstable reasoning. We propose Queryaware Subgraph Retrieval Augmented Generation (QSRAG), a retrieval framework built upon a Query-Relational Graph Attention Network (QR-GAT) that integrates query semantics and relation embeddings directly into the attention mechanism, enabling fine-grained triple scoring and scalable subgraph construction. This query-relation conditioning improves relevance estimation and suppresses noisy edges, producing faithful reasoning subgraphs. Experiments on WebQSP and CWQ establish new state-of-the-art results in both Triple Recall and Answer Recall, and significantly enhance LLMs reasoning accuracy without fine-tuning. These findings underscore the effectiveness of modeling query-relation interactions for reliable knowledge-augmented reasoning.