DeepRAG: Thinking to Retrieve Step by Step for Large Language Models

TL;DR

DeepRAG introduces a novel framework that models retrieval-augmented reasoning as an MDP, enabling adaptive retrieval and significantly improving answer accuracy in large language models.

Contribution

It proposes DeepRAG, a method that dynamically decomposes queries and decides between retrieval and parametric reasoning, enhancing retrieval efficiency and accuracy.

Findings

Improves answer accuracy by 26.4%.

Enhances retrieval efficiency in reasoning tasks.

Effectively models retrieval as an MDP for adaptive reasoning.

Abstract

Large Language Models (LLMs) have shown remarkable reasoning capabilities, while their practical applications are limited by severe factual hallucinations due to limitations in the timeliness, accuracy, and comprehensiveness of their parametric knowledge. Meanwhile, enhancing retrieval-augmented generation (RAG) with reasoning remains challenging due to ineffective task decomposition and redundant retrieval, which can introduce noise and degrade response quality. In this paper, we propose DeepRAG, a framework that models retrieval-augmented reasoning as a Markov Decision Process (MDP), enabling reasonable and adaptive retrieval. By iteratively decomposing queries, DeepRAG dynamically determines whether to retrieve external knowledge or rely on parametric reasoning at each step. Experiments show that DeepRAG improves retrieval efficiency and boosts answer accuracy by 26.4%, demonstrating its effectiveness in enhancing retrieval-augmented reasoning.