arXiv cs.CL· Nico Morloher, Johannes Reschke, Mohamed Mady·· 11 小时前AI 评分31
论文用 xAI 分析 AI 文本检测器在领域偏移下的特异性偏差与维基百科悖论
Specificity Bias and the Wikipedia Paradox: An xAI Analysis of AI Text Detectors under Domain Shift
AI 导读
一项 arXiv 研究用 SHAP 框架分析 DistilBERT、RoBERTa 等微调 Transformer 检测器在超本地新闻 OOD 数据集上的表现，揭示三种现象：检测器依赖指令微调 LLM 固有的"行动号召"特征；LLM 通过幻觉具体命名实体和模仿对话瑕疵形成"特异性偏差"绕过检测；正式客观的人类写作因低统计突发性被误判为 AI，即"维基百科悖论"。
正文 · 原文
Abstract:The widespread adoption of Large Language Models (LLMs) demands reliable detection of human-authored versus synthetic text. However, current Transformer-based detectors often experience severe performance degradation during domain shifts and operate as opaque classifiers. This paper investigates the robustness and interpretability of fine-tuned Transformer models (DistilBERT, RoBERTa) by evaluating them against a custom out-of-distribution (OOD) dataset of hyper-local news synthesized via open-source LLMs (Llama 3, Mistral, Gemma). To analyze model decision boundaries, we apply a tripartite Explainable AI (xAI) framework using SHAP. Our analysis reveals three key phenomena: First, detectors often rely on a "Call-to-Action" syndrome inherent to instruction-tuned LLMs rather than generic machine text. Second, we observe a "Specificity Bias," where LLMs successfully bypass detection by hallucinating specific named entities and mimicking conversational imperfection. Finally, we illustrate the "Wikipedia Paradox," wherein formal, objective human writing is frequently misclassified as AI due to its low statistical burstiness. These findings demonstrate that purely statistical detection paradigms face systematic limitations in highly structured human domains, emphasizing the critical need for semantically-aware, interpretable detection architectures.
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2611.03280 [cs.CL] |
| (or arXiv:2611.03280v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2611.03280 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Nico Morloher [view email]
[v1]
Mon, 2 Nov 2026 15:22:34 UTC (2,404 KB)
来源：arXiv cs.CL · arxiv.org