VikingRAG：面向结构化文档的高效检索增强生成
原标题：VikingRAG: Accurate and Token-efficient Retrieval-augmented Generation over Structured Documents
AI 摘要
该论文提出 VikingRAG，一个面向结构化文档的目录感知语义数据管理系统，将语义与结构访问紧密结合，支持基于证据缺口的多轮检索。系统通过将多轮检索轨迹物化为可复用的经验边，并引入自适应升级策略，在证据充分时仅用单轮经验增强检索。实验显示基础系统在保持高准确率的同时仅消耗现有方法 11.6%–51.9% 的 token，结合轨迹复用与自适应升级后可降至 5.1%–32.5%。
正文节选
VikingRAG: Accurate and Token-efficient Retrieval-augmented Generation over Structured Documents Abstract. State-of-the-art retrieval-augmented generation (RAG) methods exploit document structures to acquire sufficient evidence, but often incur substantial token costs. To reduce structural-context tokens without compromising high RAG accuracy, we present VikingRAG, a directory-aware semantic data management system that tightly integrates semantic and structural access to support structural-conte