Beyond Semantic Relevance: Counterfactual Risk Minimization for Robust Retrieval-Augmented Generation

2026-05-02

AI Summary

Standard retrieval-augmented generation (RAG) systems rely on semantic relevance and are prone to retrieving confirmatory evidence when users exhibit cognitive biases—such as false premises or confirmation bias—thereby exacerbating hallucinations and creating a “relevance–robustness gap.” This work proposes CoRM-RAG, the first framework to integrate counterfactual risk minimization into RAG. By applying causal interventions, it aligns retrieval with decision safety and introduces a cognitive perturbation protocol to simulate user biases. From this, a lightweight Evidence Critic module is distilled to identify documents with high evidential strength. The approach significantly outperforms existing dense retrievers and rerankers under adversarial queries and enables risk-aware abstention based on robustness scores.

Abstract

Standard Retrieval-Augmented Generation (RAG) systems predominantly rely on semantic relevance as a proxy for utility. However, this assumption collapses in realistic decision-making scenarios where user queries are laden with cognitive biases, such as false premises or confirmation bias. In such cases, maximizing relevance paradoxically promotes the retrieval of sycophantic evidence that reinforces hallucinations, a critical failure we term the ``Relevance-Robustness Gap''. To bridge this gap, we propose CoRM-RAG (Counterfactual Risk Minimization for RAG), a framework that aligns retrieval with decision safety rather than mere similarity. Grounded in causal intervention, we introduce a Cognitive Perturbation Protocol to simulate user biases during training, which is then distilled into a lightweight Evidence Critic. This scoring module learns to identify documents that possess sufficient evidential strength to steer the model toward correctness despite adversarial query perturbations. Extensive experiments on decision-making benchmarks demonstrate that CoRM-RAG significantly outperforms strong dense retrievers and LLM-based rerankers in adversarial settings, while enabling effective risk-aware abstention through reliable robustness scoring. Our code is available at https://github.com/PeiYangLiu/CoRM-RAG.git.

Problem

• Retrieval-Augmented Generation
• semantic relevance
• cognitive bias
• hallucination
• robustness

Innovation

• Counterfactual Risk Minimization
• Retrieval-Augmented Generation
• Cognitive Bias
• Causal Intervention
• Evidence Critic