CrEst: Credibility Estimation for Contexts in LLMs via Weak Supervision

TL;DR

CrEst is a weak supervision framework that assesses the credibility of context documents in LLMs by leveraging inter-document agreement, improving inference accuracy without manual annotations.

Contribution

We introduce CrEst, a novel weakly supervised method for credibility estimation in LLM contexts, utilizing semantic coherence among documents to enhance knowledge reliability.

Findings

Achieves up to 26.86% accuracy improvement

Increases F1 score by 3.49%

Maintains robustness under high-noise conditions

Abstract

The integration of contextual information has significantly enhanced the performance of large language models (LLMs) on knowledge-intensive tasks. However, existing methods often overlook a critical challenge: the credibility of context documents can vary widely, potentially leading to the propagation of unreliable information. In this paper, we introduce CrEst, a novel weakly supervised framework for assessing the credibility of context documents during LLM inference--without requiring manual annotations. Our approach is grounded in the insight that credible documents tend to exhibit higher semantic coherence with other credible documents, enabling automated credibility estimation through inter-document agreement. To incorporate credibility into LLM inference, we propose two integration strategies: a black-box approach for models without access to internal weights or activations, and a white-box method that directly modifies attention mechanisms. Extensive experiments across three model architectures and five datasets demonstrate that CrEst consistently outperforms strong baselines, achieving up to a 26.86% improvement in accuracy and a 3.49% increase in F1 score. Further analysis shows that CrEst maintains robust performance even under high-noise conditions.