Unverified90% confidenceFactTime unknown
RAG索引阶段的文本分块可以按句子分段，也可以按固定长度（如128或1024个Token）切分
1
Sources
90%
Confidence
Long-term
Relevance
6/1/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedCLIP文本编码器的标准上下文窗口长度约为77个Token，超出部分的语义信息会被截断或降权处理67% similarUnverifiedRecursive character chunking is LangChain's default approach, which sequentially attempts to split at hierarchical boundaries such as paragraphs, sentences, and words in order to prioritize semantic integrity.67% similarUnverified文本块重叠处理可在关键知识点跨越切分边界时保留语义连贯性，例如chunk_size=500、chunk_overlap=5064% similarUnverified现代LLM的分词器通常将文本压缩为约10万词汇量的Token序列64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/18822API
curl https://kongchang.com/api/v1/knowledge/claims/18822MCP
get_claim(id=18822)