README.md

metadata

license: mit
configs:
  - config_name: reddit
    default: true
    data_files: english/reddit/reddit_title_text_2021.jsonl.gz
  - config_name: s2orc
    data_files: english/s2orc/*
  - config_name: wikipedia
    data_files: english/wikipedia/GPT-wiki-intro.csv
  - config_name: wikihow
    data_files: english/wikihow/*
  - config_name: trustpilot_reviews
    data_files: english/trustpilot-reviews/trustpilot_reviews_2005.csv
  - config_name: amazon_reviews
    data_files: english/amazon-reviews/*
  - config_name: yahoo_answers
    data_files: english/yahoo-answers/*
  - config_name: natural_questions
    data_files: english/natural-questions/*
  - config_name: cc_news
    data_files: english/cc-news/*
  - config_name: npr_news
    data_files: english/npr-news/*
  - config_name: baidu_tieba
    data_files: chinese/baidu-tieba/*
  - config_name: weibo_review
    data_files: chinese/weibo-review/*
  - config_name: rednote_review
    data_files: chinese/rednote-review/*.parquet
  - config_name: csl
    data_files: chinese/csl/csl_camera_readly.tsv
  - config_name: baidu_baike
    data_files: chinese/baidu-baike/563w_baidubaike.json
  - config_name: dianping
    data_files: chinese/dianping/*
  - config_name: douban_review
    data_files: chinese/douban-review/all_movies_with_id.csv
  - config_name: baidu_zhidao
    data_files: chinese/baidu-zhidao/*
  - config_name: zhihu
    data_files: chinese/zhihu/*
  - config_name: clts
    data_files: chinese/clts/*
language:
  - en
  - zh

This is a mirror of the sourced human text datasets used for our MAGA.

Please note that this is a mirror of the original source, and some of the data in it needs to be cleaned before use. For more details, please refer to our GitHub.

domain	source
Reddit	https://huggingface.co/datasets/sentence-transformers/reddit-title-body
S2ORC	https://huggingface.co/datasets/sentence-transformers/s2orc
Wikipedia	https://huggingface.co/datasets/aadityaubhat/GPT-wiki-intro
wikiHow	https://modelscope.cn/datasets/sentence-transformers/wikihow
Trustpilot	https://huggingface.co/datasets/Kerassy/trustpilot-reviews-123k
Amazon Reviews	https://huggingface.co/datasets/sentence-transformers/amazon-reviews
Yahoo Answers	https://huggingface.co/datasets/sentence-transformers/yahoo-answers
Natural Questions	https://huggingface.co/datasets/sentence-transformers/natural-questions
CC News	https://huggingface.co/datasets/sentence-transformers/ccnews
NPR News	https://huggingface.co/datasets/sentence-transformers/npr
Baidu Tieba (百度贴吧)	https://huggingface.co/datasets/HongzheBi/DuReader2.0
Weibo Review（微博评论）	https://huggingface.co/datasets/LZYFirecn/weibo-hot-search
Rednote Review（小红书评论）	https://huggingface.co/datasets/Jingyi77/CHASM-Covert_Advertisement_on_RedNote
CSL（中文核心期刊摘要）	https://github.com/ydli-ai/CSL
Baidu Baike (百度百科)	https://huggingface.co/datasets/lars1234/baidu-baike-dataset
Dianping（大众点评）	https://huggingface.co/datasets/twn39/dianping
Douban Review (豆瓣评论)	https://huggingface.co/datasets/dirtycomputer/douban_movie_review
Baidu Zhidao (百度知道)	https://huggingface.co/datasets/HongzheBi/DuReader2.0
Zhihu（知乎）	https://huggingface.co/datasets/suolyer/zhihu
CLTS（澎湃新闻）	https://github.com/lxj5957/CLTS-Dataset

Citation

If you find MAGA useful for your research and applications, please cite using the Bibtex:

@misc{song2026maga,
      title={MAGA-Bench: Machine-Augment-Generated Text via Alignment Detection Benchmark}, 
      author={Anyang Song and Ying Cheng and Yiqian Xu and Rui Feng},
      year={2026},
      eprint={2601.04633},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2601.04633}, 
}