metadata
dataset_info:
features:
- name: text_id
dtype: string
- name: page_url
dtype: string
- name: page_title
dtype: string
- name: section_title
dtype: string
- name: context_page_description
dtype: string
- name: context_section_description
dtype: string
- name: media
sequence: string
- name: hierachy
sequence: string
- name: category
sequence: string
- name: source_id
dtype: string
splits:
- name: train
num_bytes: 14378574060.336058
num_examples: 10134744
download_size: 6408012391
dataset_size: 14378574060.336058
license: cc-by-sa-4.0
size_categories:
- 100M<n<1B
Dataset Card for "AToMiC-Texts-Mapped"
Dataset Description
- Homepage: AToMiC homepage
- Source: WIT
- Paper: WIT: Wikipedia-based Image Text Dataset for Multimodal Multilingual Machine Learning
Languages
This dataset only contains English in Wikipedia (parsed from the 20221101 XML dump).
Data Instances
Each instance is a section of a Wikipedia page. We also provide its page-level information, and associated information such as categories and media.
The source_id can be mapped back to the instance in the original WIT instance.
Notice that the WIT dataset is crawled from the earlier version of Wikipedia (2020-08-30).
The WIT dataset is mapped to the new dump by pure BM25 matching with Anserini.
Intended Usage
- Text collection for Image-to-Text retrieval
- Language model pretraining
- Document classification
Licensing Information
CC BY-SA 4.0 international license
Citation Information
TBA
Acknowledgement
Thanks to: mwparserfromhell Datasets Anserini