hannxu/hc_var

---
license: apache-2.0
task_categories:
- text-classification
language:
- en
size_categories:
- 100M<n<1B
---

# Dataset Card for HC-Var (Human and ChatGPT Texts with Variety)

This is a collection of human texts and ChatGPT (GPT3.5-Turbo) generated texts, to faciliate studies such as generated texts detection.
It includes the texts which are generated / human written to accomplish various language tasks with various approaches.
The included language tasks and topics are summarized below. Note: For each language task, this dataset considers 3 different prompts to inquire ChatGPT outputs.
The example code to train binary classification models is in [this website](https://github.com/hannxu123/hc_var).
A technical report on some representative detection methods can be find in [this paper](https://arxiv.org/abs/2310.01307).
This dataset is collected by Han Xu from Michigan State
University. Potential issues and suggestions are welcomed to be dicussed in the community panel or emails to [e-mail].

## Key variables in the dataset:
**text**: The text body (including either human or ChatGPT texts.)\
**domain**: The language tasks included in this dataset: News, Review, (Essay) Writing, QA\
**topic**: The topic in each task.\
**prompt**: The prompt used to obtain ChatGPT outputs. "N/A" for human texts.\
**pp_id**: Each task has 3 prompts to inquire ChatGPT outputs. The "pp_id" denotes the index of prompt. "0" for human texts. "1-3" for ChatGPT texts.\
**label**: "0" for human texts. "1" for ChatGPT texts.

## To cite this dataset
```
@misc{xu2023generalization,
title={On the Generalization of Training-based ChatGPT Detection Methods},
author={Han Xu and Jie Ren and Pengfei He and Shenglai Zeng and Yingqian Cui and Amy Liu and Hui Liu and Jiliang Tang},
year={2023},
eprint={2310.01307},
archivePrefix={arXiv},
primaryClass={cs.CL}
}
```

原始信息汇总

数据集卡片 for HC-Var (Human and ChatGPT Texts with Variety)

概述

HC-Var 数据集是一个包含人类文本和 ChatGPT (GPT3.5-Turbo) 生成文本的集合，旨在促进生成文本检测等研究。该数据集包括用于完成各种语言任务的生成文本和人类书写文本，涵盖多种语言任务和主题。

语言任务和主题

数据集包含以下语言任务和主题：

• 任务: 新闻、评论、（论文）写作、问答
• 主题: 每个任务的具体主题

关键变量

• text: 文本内容（包括人类或 ChatGPT 文本）
• domain: 数据集包含的语言任务：新闻、评论、（论文）写作、问答
• topic: 每个任务中的主题
• prompt: 用于获取 ChatGPT 输出的提示。人类文本为 "N/A"
• pp_id: 每个任务有 3 个提示用于获取 ChatGPT 输出。"pp_id" 表示提示的索引。人类文本为 "0"，ChatGPT 文本为 "1-3"
• label: 人类文本为 "0"，ChatGPT 文本为 "1"

引用

@misc{xu2023generalization, title={On the Generalization of Training-based ChatGPT Detection Methods}, author={Han Xu and Jie Ren and Pengfei He and Shenglai Zeng and Yingqian Cui and Amy Liu and Hui Liu and Jiliang Tang}, year={2023}, eprint={2310.01307}, archivePrefix={arXiv}, primaryClass={cs.CL} }