Unverified50% confidenceFactExact time
Parent–child retrieval uses small chunks for precise matching, then locates and returns the larger section to which the matched chunk belongs to the answer model, thereby separating 'finding the location' from 'providing context'.
1
Sources
50%
Confidence
Long-term
Relevance
10/17/2026
First Seen
Sources
Related Entities
Related Claims
Unverified父子文档检索（small-to-big）用较小的块提高检索精度，在送入LLM时回溯到更大的父级块以提供充分上下文，实现检索粒度与生成粒度的解耦63% similarUnverifiedIn synthetic data, replacing identifiers, names, section order, and surface wording can create samples that appear different but retain the same underlying phrase templates, which provides shortcuts for the model; exact duplicate detection cannot find this overlap, and finer-grained methods such as n-gram overlap, embedding similarity, or template family tracking are needed.63% similarUnverifiedTraditional ranking tracking relies on keyword positions and link crawling, whereas AI retrieval selects material through embedding vector similarity, semantic reranking, and summary synthesis, a process that produces no crawlable citation links.62% similarUnverifiedre.search() searches for a pattern at any position in the entire string and returns a result as soon as one match is found.62% similarUnverifiedTo locate the w value, two methods can be used: keyword search (e.g., searching w: to narrow the scope) or setting breakpoints from event listeners and tracing back along the call stack61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1056909API
curl https://kongchang.com/api/v1/knowledge/claims/1056909MCP
get_claim(id=1056909)