ERA: Evidence-based Reliability Alignment for Honest Retrieval-Augmented Generation

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Shin, Sunguk, Cha, Meeyoung, Lee, Byung-Jun, Park, Sungwon
Natura: Preprint
Pubblicazione: 2026
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866915950402469888
author Shin, Sunguk
Cha, Meeyoung
Lee, Byung-Jun
Park, Sungwon
author_facet Shin, Sunguk
Cha, Meeyoung
Lee, Byung-Jun
Park, Sungwon
contents Retrieval-Augmented Generation (RAG) grounds language models in factual evidence but introduces critical challenges regarding knowledge conflicts between internalized parameters and retrieved information. However, existing reliability methods, typically relying on scalar confidence, fail to explicitly distinguish between epistemic uncertainty and inherent data ambiguity in such hybrid scenarios. In this paper, we propose a new framework called ERA (Evidence-based Reliability Alignment) to enhance abstention behavior in RAG systems by shifting confidence estimation from scalar probabilities to explicit evidence distributions. Our method consists of two main components: (1) Contextual Evidence Quantification, which models internal and external knowledge as independent belief masses via the Dirichlet distribution, and (2) Quantifying Knowledge Conflict, which leverages Dempster-Shafer Theory (DST) to rigorously measure the geometric discordance between information sources. These components are used to disentangle epistemic uncertainty from aleatoric uncertainty and modulate the optimization objective based on detected conflicts. Experiments on standard benchmarks and a curated generalization dataset demonstrate that our approach significantly outperforms baselines, optimizing the trade-off between answer coverage and abstention with superior calibration.
format Preprint
id arxiv_https___arxiv_org_abs_2604_20854
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle ERA: Evidence-based Reliability Alignment for Honest Retrieval-Augmented Generation
Shin, Sunguk
Cha, Meeyoung
Lee, Byung-Jun
Park, Sungwon
Information Retrieval
Artificial Intelligence
Retrieval-Augmented Generation (RAG) grounds language models in factual evidence but introduces critical challenges regarding knowledge conflicts between internalized parameters and retrieved information. However, existing reliability methods, typically relying on scalar confidence, fail to explicitly distinguish between epistemic uncertainty and inherent data ambiguity in such hybrid scenarios. In this paper, we propose a new framework called ERA (Evidence-based Reliability Alignment) to enhance abstention behavior in RAG systems by shifting confidence estimation from scalar probabilities to explicit evidence distributions. Our method consists of two main components: (1) Contextual Evidence Quantification, which models internal and external knowledge as independent belief masses via the Dirichlet distribution, and (2) Quantifying Knowledge Conflict, which leverages Dempster-Shafer Theory (DST) to rigorously measure the geometric discordance between information sources. These components are used to disentangle epistemic uncertainty from aleatoric uncertainty and modulate the optimization objective based on detected conflicts. Experiments on standard benchmarks and a curated generalization dataset demonstrate that our approach significantly outperforms baselines, optimizing the trade-off between answer coverage and abstention with superior calibration.
title ERA: Evidence-based Reliability Alignment for Honest Retrieval-Augmented Generation
topic Information Retrieval
Artificial Intelligence
url https://arxiv.org/abs/2604.20854