arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2502.15902cs.LGcs.AIcs.CL
收藏

IPAD:用于AI检测的反向提示——一种鲁棒且可解释的LLM生成文本检测器

IPAD: Inverse Prompt for AI Detection - A Robust and Interpretable LLM-Generated Text Detector

Computer Science and Engineering, Hong Kong University of Science and Technology(香港科技大学计算机科学与工程学院);School of Computing and Data Science, The University of Hong Kong(香港大学计算机科学与数据科学学院);School of Information Science & Engineering, Lanzhou University(兰州大学信息科学与工程学院);Max Planck Institute for Security and Privacy(安全与隐私研究所);Computer Science, The University of Michigan(密歇根大学计算机科学系);Center for Data Science, Peking University(北京大学数据科学中心)

Zheng Chen, Yushi Feng, Jisheng Dang, Yue Deng, Changyang He, Hongxi Pu, Haoxuan Li, Bo Li

AI总结:

针对LLM生成文本检测中现有方法鲁棒性差、缺乏可解释性的问题,提出IPAD框架,通过提示逆变器和区分器提升分布内、分布外及受攻击数据的检测性能,并提供可解释证据增强可信度。

摘要:

大型语言模型(LLMs)在文本生成方面已达到人类级别的流畅度,这使得区分人类撰写与LLM生成的文本变得复杂。这增加了滥用风险,凸显了对可靠检测器的需求。然而,现有检测器在分布外(OOD)数据和受攻击数据上鲁棒性较差,这对现实场景至关重要。此外,它们难以提供可解释证据支持决策,从而削弱可靠性。针对这些挑战,我们提出IPAD(用于AI检测的反向提示),这是一个包含提示逆变器和两个区分器的新型框架:提示逆变器识别可能生成输入文本的预测提示,区分器检查输入文本与预测提示的对齐概率。实证评估表明,IPAD在分布内数据上比最强基线高9.05%(平均召回率),在分布外数据上高12.93%(AUROC),在受攻击数据上高5.48%(AUROC);它在结构化数据集上也表现鲁棒。此外,可解释性评估显示,IPAD通过让用户直接检查决策证据增强AI检测可信度,为其最先进的检测结果提供可解释支持。

英文摘要:

Large Language Models (LLMs) have attained human-level fluency in text generation, which complicates the distinguishing between human-written and LLM-generated texts. This increases the risk of misuse and highlights the need for reliable detectors. Yet, existing detectors exhibit poor robustness on out-of-distribution (OOD) data and attacked data, which is critical for real-world scenarios. Also, they struggle to provide interpretable evidence to support their decisions, thus undermining the reliability. In light of these challenges, we propose IPAD (Inverse Prompt for AI Detection), a novel framework consisting of a Prompt Inverter that identifies predicted prompts that could have generated the input text, and two Distinguishers that examine the probability that the input texts align with the predicted prompts. Empirical evaluations demonstrate that IPAD outperforms the strongest baselines by 9.05% (Average Recall) on in-distribution data, 12.93% (AUROC) on out-of-distribution data, and 5.48% (AUROC) on attacked data. IPAD also performs robustly on structured datasets. Furthermore, an interpretability assessment is conducted to illustrate that IPAD enhances the AI detection trustworthiness by allowing users to directly examine the decision-making evidence, which provides interpretable support for its state-of-the-art detection results.

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。

↑