[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"aihot-art-99757":3},{"itemId":4,"vertical":5,"category":6,"source":7,"score":8,"title":9,"summary":10,"analysis":11,"url":12,"coverUrl":13,"direction":13,"marketSignal":13,"publishedAt":14},"99757","ai","论文研究","面壁智能 OpenBMB",62,"清华团队提出 KG-Infused RAG：用知识图谱激活检索","清华团队提出 KG-Infused RAG，通过知识图谱扩散激活同时检索文本与事实。","清华团队提出KG-Infused RAG，将知识图谱的扩散激活机制融入检索增强生成，同时利用文本段落和结构化事实提升答案质量。\n· 方法从查询的种子实体出发，在知识图谱中逐跳扩散，由LLM选择相关三元组构建子图，避免循环。\n· 激活的事实用于扩展查询和双重检索，生成阶段将检索段落和知识图谱摘要融合为事实增强的笔记。\n· 在四个多跳数据集上，KG-Infused RAG显著优于标准RAG基线，DPO训练的LLaMA3.1-8B平均提升3.8%至13.8%。\n· 该方案即插即用，可融入Self-RAG等框架，且利用现有知识图谱，避免实时构建的噪声。\n看点：知识图谱与RAG的结合为复杂推理任务提供了新范式，有望成为下一代检索系统的标配。","https:\u002F\u002Fx.com\u002FOpenBMB\u002Fstatus\u002F2079551973024473553",null,"2026-07-21 21:00:01"]