AI vs Human Text Dataset

本数据集汇集了人类撰写文本与多种大语言模型（GPT-4、Claude-3等）生成的对偶文本，覆盖新闻、社交媒体、环境保护、生活分享、学生作文、创意写作等常见领域。每条数据包含、文本内容、生成模型标识（仅AI）、若干语言特征指标（困惑度、突发性、重复度、文本长度等）以及可选的检测难度标签。数据集主要用于AI生成文本检测模型的训练与评测、人机写作风格对比分析等。This dataset curates paired text samples, including human-written texts and texts generated by various large language models (LLMs) such as GPT-4, Claude-3, etc. It covers common domains including news, social media, environmental protection, daily life sharing, student compositions, and creative writing. Each data entry contains the text content, the generating model identifier (only for AI-generated samples), multiple linguistic feature metrics (perplexity, burstiness, repetition rate, text length, etc.), and an optional detection difficulty label. This dataset is primarily intended for the training and evaluation of AI-generated text detection models, as well as comparative analyses of writing styles between humans and AI.

搜集汇总

背景与挑战

背景概述

该数据集汇集了人类撰写文本与多种大语言模型（如GPT-4、Claude-3）生成的对偶文本，覆盖新闻、社交媒体、学生作文等多个常见领域，并包含语言特征指标和检测难度标签。它主要用于AI生成文本检测模型的训练与评估，以及人机写作风格的对比分析研究。