> 屬於一種 [[Context Enrichment 策略]] > Split and embed *small* chunks (for maximum information density), then return the parent documents (or larger chunks) those small chunks come from https://medium.com/ai-insights-cobet/rag-and-parent-document-retrievers-making-sense-of-complex-contexts-with-code-5bd5c3474a8a https://twitter.com/jerryjliu0/status/1732503009891127676 https://medium.aiplanet.com/advanced-rag-providing-broader-context-to-llms-using-parentdocumentretriever-cc627762305a ## LongRAG paper * https://arxiv.org/abs/2406.15319 (2024/6) * 將文件分組 group 去組成每個 4k 長度的文件 * 每個文件拆 child chunk 512 tokens * chunk 命中後，回傳 parent document 有 4k 長塞下 prompt 中 * 大約塞 4~8 份文件