Unverified50% confidenceFactExact time
当向量库中存在大量语义相近文档时，相似度分数趋于集中，导致Top-K召回结果区分度下降，即语义粘连问题
1
Sources
50%
Confidence
Long-term
Relevance
7/7/2026
First Seen
Sources
企业级RAG知识库落地的5大工程硬伤：从Demo到生产的真实门槛
bilibili赋范课堂7/4/2026
Related Claims
UnverifiedTop-K的K值存在经典权衡：K太小会漏掉关键信息，K太大会引入噪声导致大模型答案跑偏73% similarUnverified研究表明当上下文窗口填满时，LLM对序列首尾信息的关注度显著高于中间部分，即'迷失在中间'效应72% similarUnverifiedIn chunking strategies, chunks that are too large cause semantic focus to blur, while chunks that are too small cause context fragmentation; fixed-window, sentence-level, and semantic-boundary recursive chunking each involve trade-offs in different scenarios, with no universally optimal solution.69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/127203API
curl https://kongchang.com/api/v1/knowledge/claims/127203MCP
get_claim(id=127203)