智能体
x.com ·
@omarsar0
CorpusMap：为 agentic search 预建实体地图，答案质量提升 6.4–11.7 分、输入 token 降 34%–57%
一条 X 帖子转述 Microsoft 及合作者的论文 CorpusMap：该方法预先在文档集合中解析反复出现的实体，为每个实体生成链接到所有提及它的文档的页面，原文档保持原位。agent 读取文档后可沿实体跳转到相关文档，避免重复检索同一证据。转述称在 7 个模型、3 个基准上答案质量提升 6.4 至 11.7 分，输入 token 减少 34% 至 57%，并优于 LLM Wiki 层及另外三种导航层；地图可在不调用 LLM 的情况下构建，并随新文档到达更新。相关数据来自二手转述，需以论文原文核实。
AI推荐理由 转述 Microsoft 等机构论文 CorpusMap：预先解析文档集合中的实体并为每个实体建立链接页，agent 读文档后可沿实体跳转、避免重复检索。摘要给出 7 个模型、3 个基准上答案质量提升 6.4–11.7 分、输入 token 下降 34%–57%，并称优于 LLM Wiki 层等三种导航层，地图构建无需 LLM 调用且可增量更新。信息密度高、方向对 agentic search 有实际价值；但为二手转述，未提供论文细节与复现条件，数据需以原文为准。
78
浏览 3,419
点赞 13
评论 8
转发 2
正文
Recommended. LLM agents love structure, so it's no surprise that a corpus improves agentic search.
DAIR.AI@dair\_ai
Banger paper from Microsoft and colleagues.
If you run agents that search a large document collection, this one is worth your time.
(bookmark it)
They introduce CorpusMap, which resolves recurring entities across the collection in advance and gives each entity a page that links to every document that mentions it.
The original documents stay in place.
The agent reads a document, follows an entity to related documents, and avoids searching for the same evidence again.
Across 7 models and three benchmarks, answer quality goes up 6.4 to 11.7 points while input tokens drop 34% to 57%.
It also beats an LLM Wiki layer and three other navigation layers.
The map can be built without LLM calls and updated as new documents arrive.
Paper: [arxiv.org/abs/2609.37226](https://arxiv.org/abs/2609.37226)
Chat with Paper: [academy.dair.ai/papers/follow-…](https://academy.dair.ai/papers/follow-the-entities-a-corpus-map-for-agentic-search-2609.37226)
[🔗 View Quoted Tweet](https://x.com/dair_ai/status/2107146326639358188)