Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DualRAG: A Dual-Process Approach to Integrate Reasoning and Retrieval for Multi-Hop Question Answering

Rong Cheng, Jinyi Liu, Yan Zheng, Fei Ni, Jiazhen Du, Hangyu Mao, Fuzheng Zhang, Bo Wang, Jianye Hao

TL;DR

DualRAG presents a dual-process RAG framework that tightly couples Reasoning-augmented Querying (RaQ) with progressive Knowledge Aggregation (pKA) to handle dynamic, multi-hop reasoning. RaQ actively detects knowledge gaps and generates targeted retrieval queries, while pKA constructs a progressive knowledge outline and summarizes retrieved content to support coherent reasoning. The approach yields substantial gains in MHQA accuracy and coherence across datasets, reduces the number of retrieval iterations, and remains effective for compact models via targeted fine-tuning. This work offers a scalable, noise-resilient solution for complex reasoning tasks with practical impact on real-world information-seeking AI systems.

Abstract

Multi-Hop Question Answering (MHQA) tasks permeate real-world applications, posing challenges in orchestrating multi-step reasoning across diverse knowledge domains. While existing approaches have been improved with iterative retrieval, they still struggle to identify and organize dynamic knowledge. To address this, we propose DualRAG, a synergistic dual-process framework that seamlessly integrates reasoning and retrieval. DualRAG operates through two tightly coupled processes: Reasoning-augmented Querying (RaQ) and progressive Knowledge Aggregation (pKA). They work in concert: as RaQ navigates the reasoning path and generates targeted queries, pKA ensures that newly acquired knowledge is systematically integrated to support coherent reasoning. This creates a virtuous cycle of knowledge enrichment and reasoning refinement. Through targeted fine-tuning, DualRAG preserves its sophisticated reasoning and retrieval capabilities even in smaller-scale models, demonstrating its versatility and core advantages across different scales. Extensive experiments demonstrate that this dual-process approach substantially improves answer accuracy and coherence, approaching, and in some cases surpassing, the performance achieved with oracle knowledge access. These results establish DualRAG as a robust and efficient solution for complex multi-hop reasoning tasks.

DualRAG: A Dual-Process Approach to Integrate Reasoning and Retrieval for Multi-Hop Question Answering

TL;DR

DualRAG presents a dual-process RAG framework that tightly couples Reasoning-augmented Querying (RaQ) with progressive Knowledge Aggregation (pKA) to handle dynamic, multi-hop reasoning. RaQ actively detects knowledge gaps and generates targeted retrieval queries, while pKA constructs a progressive knowledge outline and summarizes retrieved content to support coherent reasoning. The approach yields substantial gains in MHQA accuracy and coherence across datasets, reduces the number of retrieval iterations, and remains effective for compact models via targeted fine-tuning. This work offers a scalable, noise-resilient solution for complex reasoning tasks with practical impact on real-world information-seeking AI systems.

Abstract

Multi-Hop Question Answering (MHQA) tasks permeate real-world applications, posing challenges in orchestrating multi-step reasoning across diverse knowledge domains. While existing approaches have been improved with iterative retrieval, they still struggle to identify and organize dynamic knowledge. To address this, we propose DualRAG, a synergistic dual-process framework that seamlessly integrates reasoning and retrieval. DualRAG operates through two tightly coupled processes: Reasoning-augmented Querying (RaQ) and progressive Knowledge Aggregation (pKA). They work in concert: as RaQ navigates the reasoning path and generates targeted queries, pKA ensures that newly acquired knowledge is systematically integrated to support coherent reasoning. This creates a virtuous cycle of knowledge enrichment and reasoning refinement. Through targeted fine-tuning, DualRAG preserves its sophisticated reasoning and retrieval capabilities even in smaller-scale models, demonstrating its versatility and core advantages across different scales. Extensive experiments demonstrate that this dual-process approach substantially improves answer accuracy and coherence, approaching, and in some cases surpassing, the performance achieved with oracle knowledge access. These results establish DualRAG as a robust and efficient solution for complex multi-hop reasoning tasks.

Paper Structure

This paper contains 68 sections, 12 equations, 8 figures, 17 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. RAG
  4. Iterative RAG
  5. DualRAG
  6. Framework of DualRAG
  7. Reasoning-augmented Querying (RaQ)
  8. Reasoner
  9. Entity Identifier
  10. Progressive Knowledge Aggregation
  11. Knowledge Summarizer
  12. Progressive Knowledge Outline
  13. Fine-Tuning for Compact Models
  14. Experiment
  15. Datasets and Metrics
  16. ...and 53 more sections

Figures (8)

  • Figure 1: Challenges in iterative RAG, illustrating the evolving knowledge demands in multi-hop reasoning and the three core challenges.
  • Figure 2: Overview of DualRAG, an iterative RAG framework for MHQA that combines Active Reasoning and Querying with Progressive Knowledge Aggregation.
  • Figure 3: The distribution of the average number of iterations per question for each method. Note that GenGround does not specify a clear termination criterion and always iterates up to the preset maximum limit. Therefore, its iteration count distribution is not included in the figure.
  • Figure 4: Comparison of the count of queries generated per question by DualRAG and DualRAG-FT. The DualRAG-FT produces fewer redundant queries, thereby reducing unnecessary retrieval calls.
  • Figure 5: Prompt for Reasoner
  • ...and 3 more figures