arXiv 2604.101892026-04-11cs.CL

FAITH：通过整合可信度和诚实度实现事实一致性

FAITH: Factuality Alignment through Integrating Trustworthiness and Honestness

Tsinghua University, Institute for Interdisciplinary Information Sciences（清华大学，交叉信息研究院）；Shanghai Qi Zhi Institute（上海期智研究院）；South China Normal University, School of Electronic Science and Engineering（华南师范大学，电子科学与工程学院）；Guangzhou Richstone Data Technologies Co., Ltd.（广州瑞驰数据技术有限公司）；LMU Munich & Munich Center for Machine Learning（慕尼黑大学 & 慕尼黑机器学习中心）

Xiaoning Dong, Chengyan Wu, Yajie Wen, Yu Chen, Yun Xue, Jing Zhang, Wei Xu, Bolei Ma

备注：: ACL 2026 Findings

阅读 PDF

arXiv 页面 ↗HTML 原文 ↗

AI总结：

FAITH通过整合可信度和诚实度信号，提升大语言模型的事实准确性。采用PPO算法优化奖励函数，并结合检索增强模块增强内外知识一致性。

摘要：

大型语言模型（LLMs）即使具备相应知识，也可能生成事实性不准确的内容，严重削弱其可靠性。现有方法通过在问答提示中引入不确定性来缓解这一问题，但这些数值评分缺乏语义丰富性，导致LLM无法正确理解其内部可信度和诚实度状态，从而影响事实一致性对齐。我们引入FAITH（通过整合可信度和诚实度的事实一致性框架），一种用于事实一致性对齐的后训练框架，整合了自然语言不确定性信号与外部知识。具体而言，我们通过计算LLM输出的置信度分数和语义熵，并将其映射到描述模型内部知识拥有（可信度）和回答行为（诚实度）的常识状态象限中。基于增强的数据，我们设计了一个考虑正确性和不确定性信号的奖励函数，并使用近端策略优化（PPO）算法微调LLM。为进一步缓解基础薄弱的响应，我们设计了一个检索增强模块，检索相关外部段落，提高内部和外部知识表示的一致性。在四个知识密集型基准测试中，广泛实验表明FAITH增强了LLM的事实准确性和真实性。

英文摘要：

Large Language Models (LLMs) can generate factually inaccurate content even if they have corresponding knowledge, which critically undermines their reliability. Existing approaches attempt to mitigate this by incorporating uncertainty in QA prompt during training, but these numerical scores lack the semantic richness for LLM to properly understand its internal states of trustworthiness and honestness, leading to insufficient factuality alignment. We introduce FAITH (Factuality Alignment through Integrating Trustworthiness and Honestness), a post-training framework for factuality alignment that integrates natural-language uncertainty signals with external knowledge. Specifically, we augment training datasets by computing confidence scores and semantic entropy from LLM outputs and mapping them into a knowledge state quadrant that describes the model's internal knowledge possession (trustworthiness) and answering behaviors (honestness) in natural language. Based on this enhanced data, we design a reward function that considers both correctness and uncertainty signals, and fine-tune the LLM using the Proximal Policy Optimization (PPO) algorithm. To further mitigate weakly grounded responses, we design a retrieval-augmented module that retrieves relevant external passages, improving the consistency between internal and external knowledge representations. Extensive experiments on four knowledge-intensive benchmarks demonstrate that FAITH enhances the factual accuracy and truthfulness of LLMs.

机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。