Wikipedia Summary Dataset

This is a dataset that can be used for research into machine learning and natural language processing. It contains all titles and summaries (or introductions) of English Wikipedia articles, extracted in September of 2017.

The dataset is different from the regular Wikipedia dump and different from the datasets that can be created by gensim because ours contains the extracted summaries and not the entire unprocessed page body. This could be useful if one wants to use the smaller, more concise, and more definitional summaries in their research. Or if one just wants to use a smaller but still diverse dataset for efficient training with resource constraints.

A summary or introduction of an article is everything starting from the page title up to the content outline.

The raw dataset leaves the original text structure intact. Additionally, we provide pre-processed versions.

Download

• 💾 raw.tar.gz (± 1GB; 459,081,607 words; 5,315,384 articles)
• 💾 tokenized.tar.gz (± 1GB; 533,211,092 words; 5,627,475 vocab; 5,315,384 articles)
• 💾 lowercased.tar.gz (± 1GB; 533,211,092 words; 5,172,571 vocab; 5,315,384 articles)
• 💾 without-punctuation.tar.gz (± 1GB; 461,749,888 words; 5,171,326 vocab; 5,315,384 articles)
• 💾 without-stop-words.tar.gz (± 0.8GB; 296,210,530 words; 5,171,164 vocab; 5,315,384 articles)
• 💾 stemmed.tar.gz (± 0.7GB; 296,210,530 words; 4,830,348 vocab; 5,315,384 articles)

Dataset contents

The tarbals contain two files. A .txt file and a .vocab file. The .txt file contains all the necessary data. Each line represents an article and contains both a title and a summary separated by |||. The lines are ordered by Wikipedia page_id. If you want to create a smaller test dataset, I would suggest sampling lines from the file and not splitting it directly.

Example from tokenized.txt:

Anarchism ||| Anarchism is a political philosophy that advocates self-governed societies based on voluntary…
Autism ||| Autism is a neurodevelopmental disorder characterized by impaired social interaction , impaired verbal…
Albedo ||| Albedo ( ) is a measure for reflectance or optical brightness ( Latin albedo , `` whiteness '' ) of…
…

There is also a .vocab file which contains the vocabulary and the count of each token. Example from tokenized.vocab:

, 27222735
the 25505452
. 21555700
of 16267241
in 13313133
and 12630336
a 10202887
is 7770405
…

Dataset construction

The dataset was constructed using a script that calls Wikipedia API for every page with their page_id. The correct way to construct summaries without any unwanted artifacts is constructing them by using the TextExtracts extension. So the API call we used, also uses the TextExtracts extension to create the summaries or introductions. As you can imagine, this takes quite a while.

https://en.wikipedia.org/w/api.php?format=json&maxlag=5&action=query&prop=extracts&exintro=&explaintext=&pageids=123|456|789

The actual downloading is done using download.py and stores the raw JSON output of the API in a separate folder. Afterwards the script process.py can combine all these API responses into two big files, i.e. a .txt file and a .vocab file.

Scripts to create the dataset are provided in this repository. They require a local Wikipedia installation and access to its MySQL database filled with data to get the page identifiers (page_id). You can fill a MySQL database with the Wikipedia data from the dump using MWDumper.

Additionally, we would ask you not to build the dataset using the official Wikipedia API if this is not needed, since building the dataset would require calling the API for every page and this puts strain on their public API. Please respect the maxlag=5 parameter if you use the official API en.wikipedia.org/w/api.php.

Research Publications

• Improving Word Embedding Compositionality using Lexicographic Definitions (will be published and presented at WWW ‘18)
• Improving the Compositionality of Word Embeddings Thesis PDF, Presentation PDF
• Analyzing the compositional properties of word embeddings Paper PDF

Please cite the following thesis if you use our data or code for your own research:

@mastersthesis{scheepers2017compositionality,
  author  = {Scheepers, Thijs}, 
  title   = {Improving the Compositionality of Word Embeddings},
  school  = {Universiteit van Amsterdam},
  year    = {2017},
  month   = {11},
  address = {Science Park 904, Amsterdam, Netherlands}
}

License (MIT)

Copyright 2017 Thijs Scheepers

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the “Software”), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.