analiza aiqbench
METEORA addresses critical gaps in sensitive RAG deployments by replacing opaque re-ranking with rationale-driven selection, significantly boosting precision and adversarial robustness while reducing evidence volume. The approach demonstrates that interpretability, efficiency, and security can be synergistic objectives rather than competing trade-offs, offering a practical path forward for high-stakes applications. Code availability suggests immediate utility for practitioners seeking to mitigate data poisoning risks in retrieval systems.
Ranking Free RAG: Replacing Re-ranking with Selection in RAG for Sensitive Domains
arXiv:2505.16014v5 Announce Type: replace Abstract: Retrieval-Augmented Generation (RAG) systems deployed in sensitive domains must provide interpretable evidence selection and robust safeguards against data poisoning, yet current approaches rely on opaque similarity-based retrieval with arbitrary top-k cutoffs that offer no explanation for their selections and remain vulnerable to adversarial manipulation. METEORA replaces re-ranking with rationale-driven selection via three components: a DPO-tuned LLM that generates explicit retrieval rationales, an Evidence Chunk Selection Engine (ECSE) that uses those rationales with statistical elbow detection for adaptive cutoff determination, and a Verifier LLM that filters poisoned evidence using the same rationales. Across six datasets, METEORA achieves 13.41% higher recall, 21.05% higher precision (without expansion), an 80% reduction in evidence volume, a 33.34% improvement in answer accuracy, and a 4.4x improvement in adversarial robustness. Human evaluation confirms genuine interpretability (3.64/5 confidence; 86% ground-truth agreement), demonstrating that interpretability, efficiency, and robustness are synergistic rather than competing objectives. The code is available in the GitHub repository https://github.com/YashSaxena21/METEORA
Abstract:Retrieval-Augmented Generation (RAG) systems deployed in sensitive domains must provide interpretable evidence selection and robust safeguards against data poisoning, yet current approaches rely on opaque similarity-based retrieval with arbitrary top-k cutoffs that offer no explanation for their selections and remain vulnerable to adversarial manipulation. METEORA replaces re-ranking with rationale-driven selection via three components: a DPO-tuned LLM that generates explicit retrieval rationales, an Evidence Chunk Selection Engine (ECSE) that uses those rationales with statistical elbow detection for adaptive cutoff determination, and a Verifier LLM that filters poisoned evidence using the same rationales. Across six datasets, METEORA achieves 13.41% higher recall, 21.05% higher precision (without expansion), an 80% reduction in evidence volume, a 33.34% improvement in answer accuracy, and a 4.4x improvement in adversarial robustness. Human evaluation confirms genuine interpretability (3.64/5 confidence; 86% ground-truth agreement), demonstrating that interpretability, efficiency, and robustness are synergistic rather than competing objectives. The code is available in the GitHub repository this https URL
Computation and Language (cs.CL)
arXiv:2505.16014 [cs.CL]
(or arXiv:2505.16014v5 [cs.CL] for this version)
Text integral sub licența deschisă a publicației · 168 cuvinte
Subiecte conexe
Can System-1-Style Decision Models Replace Dedicated Rerankers in RAG? A Systems Evaluation
arXiv · 27 oct. 2026
Spruce: Scalable Private Outsourced Retrieval Using Compact Embeddings
arXiv · 4 sept. 2026
Diagnosing LLM Reranker Behavior Under Fixed Evidence Pools
arXiv · 9 sept. 2026
SkillContrast: Difference-Guided Text Selection for Agent Skill Reranking
arXiv · 10 oct. 2026