ChineseWebText: Large-Scale High-quality Chinese Web Text Extracted with Effective Evaluation Model

This directory contains the ChineseWebText dataset, and the EvalWeb tool-chain to process CommonCrawl Data. Our EvalWeb tool is publicly available on github https://github.com/CASIA-LM/ChineseWebText.

ChineseWebText

• Dataset Overview

We release the latest and largest Chinese dataset ChineseWebText, which consists of 1.42 TB data and each text is assigned a quality score, facilitating LLM researchers to select data according to a new quality threshold. We also release a much Cleaner subset of 600 GB Chinese texts with quality exceeding 90% .

• Data Example

  {
      "title": "潍坊银行2021年上半年净利润同比增长29.57% 不良率降至1.10%_财经_中国网",
      "score": 0.95,
      "text": "潍坊银行2021年上半年净利润同比增长29.57% 不良率降至1.10%\n中国网财经8月24日讯 潍坊银行昨日披露2021年二季度信息报告显示，截至2021			年6月末，潍坊银行资产总额1920.44亿元，较上年末增长9.34%；负债总额1789.16亿元，较上年末增长10.54%。2021年上半年，潍坊银行实现净利润			6.09亿元，同比增长29.57%。\n资产质量方面，截至2021年6月末，潍坊银行不良贷款率1.10%，较上年末下降0.13个百分点。\n资本金方面，截至				2021年6月末，潍坊银行资本充足率、核心一级资本充足率、一级资本充足率分别为11.66%、7.89%、10.13%，分别较上年末下降1.89、0.89、1.15			个百分点。",
      "url": "http://finance.china.com.cn/news/special/2021bnb/20210824/5638343.shtml",
      "source\_domain": "finance.china.com.cn"
  }
  

  • "title": 【string】The title of the data text.
  • "score": 【float】Quality score generated by the quality evaluation model.
  • "text": 【string】Text content of data sample.
  • "url": 【string】External URL, points to the original web address of the text.
  • "source_domain": 【string】The domain name of the source website.

Citation

Please cite the paper if you use the data in this repo.

@misc{chen2023chinesewebtext,
      title={ChineseWebText: Large-scale High-quality Chinese Web Text Extracted with Effective Evaluation Model}, 
      author={Jianghao Chen and Pu Jian and Tengxiao Xi and Dongyi Yi and Qianlong Du and Chenglin Ding and Guibo Zhu and Chengqing Zong and Jinqiao Wang and Jiajun Zhang},
      year={2023},
      eprint={2311.01149},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}