DeepNote: Note-Centric Deep Retrieval-Augmented Generation

TL;DR

DeepNote introduces a note-centric adaptive retrieval framework for RAG that enhances knowledge exploration and improves answer accuracy in question-answering tasks by leveraging notes to refine retrieval and generation.

Contribution

It presents a novel note-centric adaptive retrieval method that better reflects information needs and fully utilizes retrieved knowledge in RAG systems.

Findings

DeepNote outperforms baselines by 10.2% to 20.1% in QA accuracy.

It effectively gathers knowledge with high density and quality.

DPO further enhances DeepNote's performance.

Abstract

Retrieval-Augmented Generation (RAG) mitigates factual errors and hallucinations in Large Language Models (LLMs) for question-answering (QA) by incorporating external knowledge. However, existing adaptive RAG methods rely on LLMs to predict retrieval timing and directly use retrieved information for generation, often failing to reflect real information needs and fully leverage retrieved knowledge. We develop DeepNote, an adaptive RAG framework that achieves in-depth and robust exploration of knowledge sources through note-centric adaptive retrieval. DeepNote employs notes as carriers for refining and accumulating knowledge. During in-depth exploration, it uses these notes to determine retrieval timing, formulate retrieval queries, and iteratively assess knowledge growth, ultimately leveraging the best note for answer generation. Extensive experiments and analyses demonstrate that DeepNote significantly outperforms all baselines (+10.2% to +20.1%) and exhibits the ability to gather knowledge with both high density and quality. Additionally, DPO further improves the performance of DeepNote. The code and data are available at https://github.com/thunlp/DeepNote.