Unverified50% confidenceConceptExact time
Once text is divorced from its original structural context, its semantics collapse, manifesting as text that should be distinguishable obtaining nearly identical vectors in the embedding space
1
Sources
50%
Confidence
Long-term
Relevance
10/13/2026
First Seen
Sources
Related Entities
Related Claims
Verifiedtext-embedding-ada-002将文本转化为1536维的稠密向量，语义相近的文本在向量空间中距离更近75% similarUnverifiedThe recursive character text splitter balances the contradiction between document chunks being too long to exceed the context window and too short to lose semantics by attempting to split at natural boundaries (paragraphs, sentences).75% similarUnverifiedText chunking granularity that is too large results in retrieved text containing large amounts of irrelevant information, while granularity that is too small may lose contextual semantics.74% similarUnverifiedTraditional frame-by-frame OCR can only detect spatial drift of text, and is nearly insensitive to the temporal instability of text content drifting across frames.67% similarUnverifiedContextual embedding introduces surrounding text or full-document information as conditional input when generating vectors, causing the same passage to produce different vector representations in different articles.67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1019556API
curl https://kongchang.com/api/v1/knowledge/claims/1019556MCP
get_claim(id=1019556)