wikipedia_subsets / README.md
UmarAzam's picture
Added Reference for dataset generation
72f5c2f verified
---
license: mit
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
dataset_info:
features:
- name: text
dtype: string
- name: hits
dtype: int64
splits:
- name: train
num_bytes: 4041398525
num_examples: 219846
download_size: 2293791651
dataset_size: 4041398525
task_categories:
- feature-extraction
language:
- en
pretty_name: 'wikipedia industrial technical '
size_categories:
- 100K<n<1M
---
This dataset was generated by filtering a subset from the wikipedia dataset -> "wikimedia/wikipedia" -> " 20231101.en"
Detailed information on how this was accomplished is given in this notebook. https://github.com/Umar-Azam/embedding_finetuner_wiki/tree/main
Short Explanation : We have a list of keywords to check against. Each wikipedia text is tokenized into word sets and the "hits" value contains the number of our filter keywords present in each text. Only the items with >4 matches are then filtered to generate this dataset.