metadata

annotations_creators:
  - no-annotation
language_creators:
  - other
languages:
  - ca
  - en
  - de
  - es
  - fr
  - ru
  - ja
  - it
  - zh
  - pt
  - nl
  - tr
  - pl
  - vi
  - ar
  - id
  - uk
  - ro
  - 'no'
  - th
  - sv
  - el
  - fi
  - he
  - da
  - cs
  - ko
  - fa
  - hi
  - hu
  - sk
  - lt
  - et
  - hr
  - is
  - lv
  - ms
  - bg
  - sr
  - ca
licenses:
  - cc0-1.0
multilinguality:
  - multilingual
pretty_name: MQA - a Multilingual FAQ and CQA Dataset
size_categories:
  - unknown
source_datasets:
  - original
task_categories:
  - question-answering
task_ids:
  - multiple-choice-qa

MQA

MQA is a Multilingual corpus of Questions and Answers (MQA) parsed from the Common Crawl. Questions are divided in two types: Frequently Asked Questions (FAQ) and Community Question Answering (CQA).

from datasets import load_dataset
all_data = load_dataset("clips/mqa", language="en")
{
  "name": "the title of the question (if any)",
  "text": "the body of the question (if any)",
  "answers": [{
    "text": "the text of the answer",
    "is_accepted": "true|false"
  }]
}
faq_data = load_dataset("clips/mqa", scope="faq", language="en")
cqa_data = load_dataset("clips/mqa", scope="cqa", language="en")

Languages

We collected around 234M pairs of questions and answers in 39 languages. To download a language specific subset you need to specify the language key as configuration. See below for an example.

load_dataset("clips/mqa", language="en") # replace "en" by any language listed below

Language	FAQ Questions	CQA Questions
en	130,344,204	7,804,182
de	12,393,508	607,436
es	10,469,289	606,335
fr	9,308,231	1,035,504
ru	8,917,939	1,150,006
it	4,748,047	307,368
ja	4,718,682	1,219,364
zh	4,596,792	410,451
pt	3,912,171	308,025
nl	3,443,730	281,342
pl	2,880,754	43,526
tr	2,845,351	200,334
vi	2,030,513	73,755
ar	1,664,424	606,190
id	1,564,936	145,973
uk	1,412,468	20,534
ro	1,344,766	76,781
no	1,270,541	9,247
th	1,185,354	23,261
sv	1,128,790	11,376
el	1,123,855	2,348
fi	993,789	20,065
da	956,871	8,805
he	935,144	53,764
ko	789,772	37,290
cs	758,390	102,013
fa	631,084	87,088
hu	496,215	12,350
sk	471,196	3,795
hi	462,052	145,906
lt	397,695	155
et	286,187	411
hr	221,410	12,846
is	219,771	12
lv	213,524	57
ms	165,703	7,259
bg	114,842	3,323
sr	87,684	8,344
ca	66,249	812

FAQ vs. CQA

You can download the Frequently Asked Questions (FAQ) or the Community Question Answering (CQA) part of the dataset.

faq = load_dataset("clips/mqa", scope="faq")
cqa = load_dataset("clips/mqa", scope="cqa")
all = load_dataset("clips/mqa", scope="all")

Although FAQ and CQA questions share the same structure, CQA questions can have multiple answers for a given questions, while FAQ questions have a single answer. FAQ questions typically only have a title (name key), while CQA have a title and a body (name and text).

Nesting and Data Fields

You can specify three different nesting level: question, page and domain.

Question

load_dataset("clips/mqa", level="question") # default

The default level is the question object:

name: the title of the question(if any) in markdown format
text: the body of the question (if any) in markdown format
answers: a list of answers
- text: the title of the answer (if any) in markdown format
- name: the body of the answer in markdown format
- is_accepted: true if the answer is selected.

Page

This level returns a list of questions present on the same page. This is mostly useful for FAQs since CQAs already have one question per page.

load_dataset("clips/mqa", level="page")

Domain

This level returns a list of pages present on the web domain. This is a good way to cope with FAQs duplication by sampling one page per domain at each epoch.

load_dataset("clips/mqa", level="domain")

Source Data

This section was adapted from the source data description of OSCAR

Common Crawl is a non-profit foundation which produces and maintains an open repository of web crawled data that is both accessible and analysable. Common Crawl's complete web archive consists of petabytes of data collected over 8 years of web crawling. The repository contains raw web page HTML data (WARC files), metdata extracts (WAT files) and plain text extracts (WET files). The organisation's crawlers has always respected nofollow and robots.txt policies.

To construct MQA, we used the WARC files of Common Crawl.

People

This model was developed by Maxime De Bruyn, Ehsan Lotfi, Jeska Buhmann and Walter Daelemans.

Licensing Information

These data are released under this licensing scheme.
We do not own any of the text from which these data has been extracted.
We license the actual packaging of these data under the Creative Commons CC0 license ("no rights reserved") http://creativecommons.org/publicdomain/zero/1.0/

Should you consider that our data contains material that is owned by you and should therefore not be reproduced here, please:
* Clearly identify yourself, with detailed contact data such as an address, telephone number or email address at which you can be contacted.
* Clearly identify the copyrighted work claimed to be infringed.
* Clearly identify the material that is claimed to be infringing and information reasonably sufficient to allow us to locate the material.

We will comply to legitimate requests by removing the affected sources from the next release of the corpus.

Citation information

@inproceedings{de-bruyn-etal-2021-mfaq,
    title = "{MFAQ}: a Multilingual {FAQ} Dataset",
    author = "De Bruyn, Maxime  and
      Lotfi, Ehsan  and
      Buhmann, Jeska  and
      Daelemans, Walter",
    booktitle = "Proceedings of the 3rd Workshop on Machine Reading for Question Answering",
    month = nov,
    year = "2021",
    address = "Punta Cana, Dominican Republic",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2021.mrqa-1.1",
    pages = "1--13",
}