DeepRAG, a combination of DeepSeek R1 hierarchical decomposition and RAG-Gym process supervision with UMLS concept rewards, reports EM 62.4 and concept accuracy 71.8 on the MedHopQA dev set.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
DeepRAG: Integrating Hierarchical Reasoning and Process Supervision for Biomedical Multi-Hop QA
DeepRAG, a combination of DeepSeek R1 hierarchical decomposition and RAG-Gym process supervision with UMLS concept rewards, reports EM 62.4 and concept accuracy 71.8 on the MedHopQA dev set.