jd5697/wikipedia-biology

Dataset Card for wikipedia-biology

Dataset Description

• —Repository: https://github.com/JaiDoshi/Knowledge-Erasure
• —Paper: https://arxiv.org/abs/2411.12103

Dataset Summary

The dataset consists of text from 87045 Wikipedia articles created by processing all articles in the Wikipedia categories Branches of biology, Biological concepts, Eukaryote biology and Biology terminology, as well as their subcategories recursively till a depth of 4. It was originally created for the purpose of unlearning the domain of biology, although it may be used for other purposes such as biology fine-tuning.

It also comprises of text from 19675 articles from three other categories: History, Concepts in physics and Philosophy, created by recursive processing till a depth of 3, to be used as the retain set in unlearning.

Note that articles such as biographies of biologists are also included.

Dataset Structure

The dataset consists of .txt files corresponding to the text in the articles. The files for the Biology dataset are contained in the subfolders under the Wikipedia Biology folder. The files for the retain set are contained in the subfolders under the Wikipedia History_Concepts-in-Physics_Philosophy folder. The files have been split into multiple subfolders due to the limit of 10k files per folder.

Licensing Information

Released under CC BY-SA 4.0 (same as Wikipedia).

Citation Information

@misc{doshi2025doesunlearningtrulyunlearn,
      title={Does Unlearning Truly Unlearn? A Black Box Evaluation of LLM Unlearning Methods}, 
      author={Jai Doshi and Asa Cooper Stickland},
      year={2025},
      eprint={2411.12103},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2411.12103}, 
}

Contributions

Thanks to @JaiDoshi for adding this dataset.