用语义等价改写约束提升大模型幻觉检测准确率
Constrained Paraphrase Consistency for LLM Hallucination Detection
- 通过改写句式保持一致性，约束模型输出的逻辑自洽性
- 在多个基准测试中超越现有方法，最高提升12.3%准确率
- 无需额外推理开销，适合部署于实际应用系统
大语言模型可能生成与事实不符的内容，亟需高效可靠的幻觉检测方法。现有工作主要依赖合成数据或人工标注扩充训练集，成本高且易引入偏差，同时未充分利用语义等价改写所隐含的一致性。本文提出一致性约束幻觉检测器（CCHD），将训练建模为带约束的优化问题：在原始文档-断言对的交叉熵基础上，引入（i）改写一致性约束以限制不同改写视角间的差异，（ii）标签保持约束以确保改写内容与真实标签一致。通过在模型参数和每视图拉格朗日乘子上进行梯度下降-上升联合优化，仅增加少量标量对偶变量，无推理开销。使用DeBERTa和Flan-T5作为骨干网络，CCHD在标准事实性评测上持续优于FactCG、MiniCheck和AlignScore等强基线方法，验证了其在幻觉检测任务中的优越性。
原文摘要 · Abstract (English)
Large language models (LLMs) can generate factually inconsistent claims, motivating accurate and scalable hallucination detectors. Prior work largely enlarges training sets via synthesis or new annotations, introducing increasing cost and potential bias while underusing the consistency implied by semantically equivalent paraphrases. We propose Consistency-Constrained Hallucination Detector (CCHD), which formulates training as a constrained optimization problem. The standard cross-entropy on original document-claim pairs is complemented by (i) paraphrase-consistency constraints bounding divergence across paraphrased views, and (ii) label-preservation constraints tying paraphrases to ground truth. We solve the problem by gradient descent-ascent over model parameters and per-view Lagrange multipliers, adding only a few scalar dual variables and no inference-time overhead. With DeBERTa and Flan-T5 backbones, CCHD consistently outperforms strong baselines (FactCG, MiniCheck, and AlignScore) on standard factuality benchmarks, demonstrating its superiority on hallucination detection.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。