Unverified50% confidenceOpinionExact time
向量检索对多跳推理（Multi-hop Reasoning）天然乏力，当答案需要综合多个文档中分散的信息片段时，单纯相似度排序往往无法检索出所有必要上下文
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/2/2026
First Seen
Valid until: 9/30/2026
Sources
吴恩达新课：用知识图谱增强RAG的完整指南
bilibili吴恩达Agentic6/11/2026
Related Claims
Unverified当知识库规模超过百万文档量级时RAG检索精度会明显衰退，对跨多文档复合推理问题单次检索往往力不从心72% similarUnverified先前的难负样本挖掘工作大多忽略非干扰项，只关注与查询的相似度，因而更易误选假负样本67% similarUnverifiedIn multi-path sampling, wrong answers tend to be sporadic rather than consistent, allowing them to be filtered out through multiple sampling67% similarUnverified传统RAG方案的核心局限包括：文档切片破坏知识整体语义结构、向量相似度检索难以捕捉深层逻辑关联、每次查询相互独立无法沉淀结构化知识67% similarUnverified知识库中存在相互矛盾的文档描述时，检索系统无法自动裁决，会导致模型生成前后不一致的答案66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/49213API
curl https://kongchang.com/api/v1/knowledge/claims/49213MCP
get_claim(id=49213)