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

TL;DR

DeepRAG is a novel framework that combines hierarchical question decomposition and process supervision to improve biomedical multi-hop question answering, achieving significant performance gains on the MedHopQA dataset.

Contribution

It introduces a unified approach integrating DeepSeek and RAG Gym with UMLS-informed supervision for enhanced biomedical QA.

Findings

Outperforms baseline models on MedHopQA

Achieves higher Exact Match scores

Improves concept level accuracy

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.