Structured Knowledge Representation through Contextual Pages for Retrieval-Augmented Generation
Structured Knowledge Representation through Contextual Pages for Retrieval-Augmented Generation [53.8] PAGERは、RAGのためのページ駆動の自律的知識表現フレームワークである。 関連文書を反復的に検索して洗練し、各スロットをポップアップさせ、最終的にコヒーレントなページを構成する。 実験の結果、PAGERはすべてのRAGベースラインを一貫して上回っている。 論文参考訳（メタデータ） (Wed, 14 Jan 2026 11:44:31 GMT)]a
「PAGER first prompts the LLM to draw on its parametric knowledge to con- struct a structured cognitive outline for the target question. This outline consists of multiple slots, each representing a distinct aspect of the potentially relevant knowledge needed to answer the question. Then PAGER employs an iterative knowledge completion mechanism to iteratively retrieve supporting documents for each slot, refine them into concise knowledge evidence, and fill the corresponding slot in the page. This iterative process continues until all slots are filled with the corresponding knowledge evidence. Finally, PAGER uses this structured page as contextual knowledge to guide the LLM to answer the given question」というフレームワークの提案。Deep Researchのような動き。