RAR$^2$: Retrieval-Augmented Medical Reasoning via Thought-Driven Retrieval

2025-09-24

AI Summary

Existing RAG methods struggle with complex biomedical questions requiring deep reasoning, as they rely on surface-level inputs and fail to accurately identify implicit knowledge needs. To address this, we propose Thought-RAG—a novel framework that jointly optimizes *reasoning-enhanced retrieval* and *retrieval-enhanced reasoning*. It explicitly models implicit chains of thought to uncover underlying clinical knowledge requirements, thereby guiding precise retrieval. Our approach integrates a thought-driven retrieval mechanism, hybrid preference pair construction, and Direct Preference Optimization (DPO) for training, augmented by test-time scaling to expand performance boundaries. Evaluated across multiple biomedical QA benchmarks, Thought-RAG significantly outperforms conventional RAG methods—achieving state-of-the-art results without task-specific fine-tuning. This demonstrates its strong generalizability, robustness, and practical applicability in real-world clinical reasoning scenarios.

Technology Category

• Knowledge Representation and Reasoning: Computational Complexity of Reasoning
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning
• Search and Optimization: Metareasoning and Metaheuristics

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Large Language Models (LLMs) have shown promising performance on diverse medical benchmarks, highlighting their potential in supporting real-world clinical tasks. Retrieval-Augmented Generation (RAG) has emerged as a key approach for mitigating knowledge gaps and hallucinations by incorporating external medical information. However, RAG still struggles with complex medical questions that require intensive reasoning, as surface-level input often fails to reflect the true knowledge needs of the task. Existing methods typically focus on refining queries without explicitly modeling the reasoning process, limiting their ability to retrieve and integrate clinically relevant knowledge. In this work, we propose RAR$^2$, a joint learning framework that improves both Reasoning-Augmented Retrieval and Retrieval-Augmented Reasoning. RAR$^2$ constructs a thought process to uncover implicit knowledge requirements and uses it to guide retrieval and answer generation. We build a training dataset of mixed preference pairs and apply Direct Preference Optimization (DPO) to train the model. Moreover, we design two test-time scaling strategies to explore the boundaries of our framework. Experiments demonstrate the effectiveness of RAR$^2$ across several biomedical question answering datasets, outperforming RAG baselines with or without fine-tuning.

Problem

• Improves reasoning-augmented retrieval for complex medical questions
• Models thought processes to uncover implicit knowledge requirements
• Enhances retrieval-augmented reasoning through joint learning framework

Innovation

• Thought-driven retrieval uncovers implicit knowledge needs
• Joint learning framework enhances retrieval and reasoning
• Direct Preference Optimization trains model with mixed preference pairs