DecEx-RAG: Boosting Agentic Retrieval-Augmented Generation with Decision and Execution Optimization via Process Supervision
This work addresses performance and efficiency bottlenecks in agentic RAG for complex task processing, offering an incremental improvement over existing methods.
The paper tackles inefficiencies in agentic retrieval-augmented generation (RAG) by proposing DecEx-RAG, which models RAG as a Markov Decision Process with decision-making and execution optimization, resulting in a 6.2% average performance improvement across six datasets and a 6x efficiency gain in data construction.
Agentic Retrieval-Augmented Generation (Agentic RAG) enhances the processing capability for complex tasks through dynamic retrieval and adaptive workflows. Recent advances (e.g., Search-R1) have shown that outcome-supervised reinforcement learning demonstrate strong performance. However, this approach still suffers from inefficient exploration, sparse reward signals, and ambiguous global reward feedback. To address these challenges, we propose DecEx-RAG, which models RAG as a Markov Decision Process (MDP) incorporating decision-making and execution, while introducing an efficient pruning strategy to optimize data expansion. Through comprehensive process-level policy optimization, DecEx-RAG significantly enhances the autonomous task decomposition, dynamic retrieval, and high-quality answer generation capabilities of large language models (LLMs). Experiments show that DecEx-RAG achieves an average absolute performance improvement of $6.2\%$ across six datasets, significantly outperforming existing baselines. Moreover, the pruning strategy improves data construction efficiency by nearly $6 \times$, providing an efficient solution for process-supervised RAG training. The code is available at https://github.com/sdsxdxl/DecEx-RAG.