Stateful Evidence-Driven Retrieval-Augmented Generation with Iterative Reasoning
For practitioners using RAG in question answering, this work offers a more robust and stable approach to evidence accumulation, though it is an incremental improvement over existing multi-step RAG methods.
The paper tackles the problem of unstable performance in Retrieval-Augmented Generation (RAG) due to flat context and stateless retrieval. The proposed framework, Stateful Evidence-Driven RAG with Iterative Reasoning, achieves consistent improvements over standard RAG and multi-step baselines on multiple QA benchmarks, while maintaining stable performance under substantial retrieval noise.
Retrieval-Augmented Generation (RAG) grounds Large Language Models (LLMs) in external knowledge but often suffers from flat context representations and stateless retrieval, leading to unstable performance. We propose Stateful Evidence-Driven RAG with Iterative Reasoning, a framework that models question answering as a progressive evidence accumulation process. Retrieved documents are converted into structured reasoning units with explicit relevance and confidence signals and maintained in a persistent evidence pool capturing both supportive and non-supportive information. The framework performs evidence-driven deficiency analysis to identify gaps and conflicts and iteratively refines queries to guide subsequent retrieval. This iterative reasoning process enables stable evidence aggregation and improves robustness to noisy retrieval. Experiments on multiple question answering benchmarks demonstrate consistent improvements over standard RAG and multi-step baselines, while effectively accumulating high-quality evidence and maintaining stable performance under substantial retrieval noise.