metadata

license: cc-by-nc-3.0
task_categories:
  - question-answering
language:
  - en
  - ru
  - pt
  - it
  - es
  - fr
  - de
  - nl
pretty_name: multilingual-wikihow-qa-16k
size_categories:
  - 10K<n<100K
tags:
  - wikihow
  - QnA
dataset_info:
  features:
    - name: INSTRUCTION
      dtype: string
    - name: RESPONSE
      dtype: string
    - name: SOURCE
      dtype: string
    - name: METADATA
      dtype: string
  splits:
    - name: train
      num_bytes: 144407512
      num_examples: 16822
  download_size: 76391535
  dataset_size: 144407512

Dataset Card for multilingual WikiHow with ~16.8K entries. ~(2-2.2)K for each language.

Warning [1]

The WikiHow team contacted me and made it clear that they forbid the use of their data for machine learning purposes. However, I am not calling for anything, and this dataset only shows the concept, and I strongly advise against violating their ToS. However, consultation with lawyers made it clear that dataset can be used for such purposes if the project has research purposes.

Warning [2]

Source code is kinda very bad, and I'm lazy to fix it.

Dataset Summary

Contains Parquet of a list of instructions and WikiHow articles on different languages.

Each row consists of

INSTRUCTION
RESPONSE
SOURCE (*.wikihow.com)
METADATA (json with url and language).

Licensing Information

Data is from WikiHow, license for content is located here: https://www.wikihow.com/wikiHow:Creative-Commons

Acknowledgements

This helped me a lot!

https://github.com/HelloChatterbox/PyWikiHow; https://pypi.org/project/pywikihow/