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.
Abstract (translated)
上下文信息的整合已显著提升了大型语言模型（LLM）在知识密集型任务上的表现。然而，现有方法往往忽视了一个关键挑战：上下文文档的可信度可能存在巨大差异，这可能导致不实信息的传播。本文介绍了CrEst，这是一种新颖的弱监督框架，在LLM推理过程中评估上下文文档的可信度——无需手动标注。我们的方法基于这样一个观点：可信的文档通常与其他可信文档在语义上更为一致，从而可以通过文档间的一致性实现自动化的可信度估计。 为了将可信度整合到LLM推理中，我们提出了两种集成策略：一种是黑盒方法，适用于无法访问内部权重或激活值的情况；另一种是白盒方法，直接修改注意力机制。在三种模型架构和五个数据集上的广泛实验表明，CrEst始终优于强大的基线方法，在准确性方面最多提高了26.86%，F1分数则提升了3.49%。进一步的分析显示，即使在高噪声条件下，CrEst也能保持其稳健的表现。
URL
https://arxiv.org/abs/2506.14912
