DeepRAG: Integrating Hierarchical Reasoning and Process Supervision for Biomedical Multi-Hop QA

2025-05-31

AI Summary

To address the challenges of complex query parsing and low domain-specific accuracy in biomedical multi-hop question answering (MedHopQA), this paper proposes a retrieval-augmented generation framework that synergistically combines hierarchical decomposition and process-level supervision. Methodologically: (1) a hierarchical question decomposer, built upon DeepSeek, explicitly models multi-hop reasoning paths; (2) RAG Gym is integrated to enable joint optimization of retrieval and generation; (3) for the first time, UMLS ontology-driven concept-level reward signals are introduced, enabling fine-grained semantic alignment and process supervision via reinforcement learning. Experiments on MedHopQA demonstrate substantial improvements over DeepSeek-V2 and RAG Gym baselines: +12.7% in Exact Match and +15.3% in concept-level accuracy. These results validate the framework’s effectiveness in precise biomedical semantic modeling and interpretable, stepwise reasoning.

Technology Category

• Natural Language Processing: Question Answering
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning
• Knowledge Representation and Reasoning: Ontologies

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

We propose DeepRAG, a novel framework that integrates DeepSeek hierarchical question decomposition capabilities with RAG Gym unified retrieval-augmented generation optimization using process level supervision. Targeting the challenging MedHopQA biomedical question answering task, DeepRAG systematically decomposes complex queries into precise sub-queries and employs concept level reward signals informed by the UMLS ontology to enhance biomedical accuracy. Preliminary evaluations on the MedHopQA dataset indicate that DeepRAG significantly outperforms baseline models, including standalone DeepSeek and RAG Gym, achieving notable improvements in both Exact Match and concept level accuracy.

Problem

• Enhances biomedical multi-hop QA accuracy
• Decomposes complex queries into precise sub-queries
• Improves retrieval-augmented generation with process supervision

Innovation

• Hierarchical question decomposition for precise sub-queries
• Process supervision with UMLS ontology rewards
• Unified RAG optimization via RAG Gym