SHIELD
收藏资源简介:
SHIELD数据集由休斯顿大学的研究团队创建,旨在为大型语言模型生成的文本检测提供一个公平且实用的评估标准。数据集包含多种写作风格,如新闻、评论、问答等,并提供了人类撰写和LLM生成的文本样本,以及针对不同难度级别的文本样本。该数据集通过引入控制难度参数的人性化框架,使得AI生成的文本更接近人类作者的写作风格,从而为检测模型的鲁棒性和稳定性提供更具挑战性的评估场景。
The SHIELD dataset was developed by a research team from the University of Houston, aiming to provide a fair and practical evaluation benchmark for detecting texts generated by large language models (LLMs). The dataset includes multiple writing styles such as news articles, product reviews, question-answering contents, etc., and provides text samples written by humans, generated by LLMs, as well as samples across different difficulty levels. This dataset introduces a human-centric framework with controllable difficulty parameters, enabling AI-generated texts to closely align with the writing styles of human authors, thus creating more challenging evaluation scenarios for testing the robustness and stability of text detection models.
SHIELD-Benchmark数据集概述
基本信息
- 数据集名称:SHIELD-Benchmark
- 当前状态:待更新(This repository will be updated.)
备注
- 该数据集目前尚未提供具体信息,建议持续关注更新。




