Stable-RAG: Mitigating Retrieval-Permutation-Induced Hallucinations in Retrieval-Augmented Generation

2026-01-06

arXiv.org

AI Summary

This work addresses the sensitivity of retrieval-augmented generation (RAG) to the order of retrieved documents, a phenomenon that often leads large language models to produce inconsistent or hallucinated outputs—even when all relevant documents are correctly retrieved. The study systematically identifies this ordering instability for the first time and proposes a stabilization mechanism based on clustering reasoning patterns. Specifically, the generator is run under multiple document orderings, and its hidden states are clustered to identify dominant reasoning paths; decoding from the cluster centroids then yields consistent, aligned outputs that effectively correct hallucinations. Evaluated on three question-answering benchmarks, the method significantly improves answer accuracy and reasoning consistency, while demonstrating strong robustness across diverse retrievers, input lengths, and cross-dataset settings.

Technology Category

• Knowledge Representation and Reasoning: Common-Sense Reasoning
• Search and Optimization: Metareasoning and Metaheuristics
• Reasoning under Uncertainty: Other Foundations of Reasoning under Uncertainty

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Retrieval-Augmented Generation (RAG) has become a key paradigm for reducing factual hallucinations in large language models (LLMs), yet little is known about how the order of retrieved documents affects model behavior. We empirically show that under Top-5 retrieval with the gold document included, LLM answers vary substantially across permutations of the retrieved set, even when the gold document is fixed in the first position. This reveals a previously underexplored sensitivity to retrieval permutations. Although robust RAG methods primarily focus on enhancing LLM robustness to low-quality retrieval and mitigating positional bias to distribute attention fairly over long contexts, neither approach directly addresses permutation sensitivity. In this paper, we propose Stable-RAG, which exploits permutation sensitivity estimation to mitigate permutation-induced hallucinations. Stable-RAG runs the generator under multiple retrieval orders, clusters hidden states, and decodes from a cluster-center representation that captures the dominant reasoning pattern. It then uses these reasoning results to align hallucinated outputs toward the correct answer, encouraging the model to produce consistent and accurate predictions across document permutations. Experiments on three QA datasets show that Stable-RAG significantly improves answer accuracy, reasoning consistency and robust generalization across datasets, retrievers, and input lengths compared with baselines.

Problem

• Retrieval-Augmented Generation
• hallucination
• permutation sensitivity
• document order
• LLM robustness

Innovation

• Retrieval-Augmented Generation
• Permutation Sensitivity
• Hallucination Mitigation
• Stable-RAG
• Reasoning Consistency