“Sounds a little like this recent paper;”
RAPTOR: Recursive Abstractive Processing for Tree-Organized Retrieval
递归地对文本块嵌入、聚类、摘要成树,检索时跨层次整合长文档信息。
多数检索增强方法只检索较短的连续文本块,不利于整体理解文档。RAPTOR 自底向上递归地对文本块做嵌入、聚类和摘要,构建出带有不同摘要层次的树;推理时从这棵树中检索,整合长文档里不同抽象层次的信息。受控实验显示它相对传统检索增强语言模型有明显提升;与 GPT-4 结合后,在 QuALITY 基准上把最佳成绩提高了绝对 20% 的准确率。
评论者在讨论 RAG 如何处理长文档、是否用 LLM 对聚类后的段落做摘要时引用它,作为分层、树状检索能带来提升的代表做法。
HN 书评
推荐在前,同类按时间 · 每篇最多保留 20 条,统计用全部提及“RAPTOR was just published and is close to SoTA, and from a quick read I can already think of several directions to improve it”
“It introduces a novel approach of recursively embedding, clustering, and summarizing chunks of text”
“Also the code for Raptor came out recently and they are showing gains through a tree/hierarchy too.”
“Instead, using LLM to generate a summary based on the clustering of paragraphs could be a good alternative. That's what RAPTOR has suggested.”
“RAPTOR: Recursive Abstractive Processing for Tree-Organized Retrieval [3]”
看全部 7 条 →
“RAPTOR for example is a technique that groups and clusters documents together, summarizes them, and creates embeddings defining a sort of a Tree.”