README.md

metadata

annotations_creators:
  - private
language_creators: null
language:
  - fr-FR
license:
  - cc-by-nc-sa-4.0
multilinguality:
  - monolingual
size_categories:
  - 1K<n<10K
source_datasets:
  - original
  - newspaper
  - online
task_categories:
  - question-answering
task_ids:
  - extractive-qa
  - open-domain-qa
paperswithcode_id: null

Dataset Card for newsquadfr

Table of Contents

• Dataset Description
  • Dataset Summary
  • Supported Tasks
  • Languages
• Dataset Structure
  • Data Instances
  • Data Fields
  • Data Splits
• Dataset Creation
  • Curation Rationale
  • Source Data
  • Annotations
  • Personal and Sensitive Information
• Considerations for Using the Data
  • Social Impact of Dataset
  • Discussion of Biases
  • Other Known Limitations
• Additional Information
  • Dataset Curators
  • Licensing Information
  • Citation Information

Dataset Description

• Homepage: lincoln.fr
• Repository: github/Lincoln-France
• Paper: [Needs More Information]
• Leaderboard: [Needs More Information]
• Point of Contact: email

Dataset Summary

newsquadfr is a small dataset created for Question Answering task. Contexts are paragraphs of articles extracted from nine online french newspaper during year 2020/2021. newsquadfr stands for Newspaper question answering dataset in french. inspired by Piaf and Squad dataset. 2 520 triplets context - question - answer.

from datasets import load_dataset

ds_name = 'lincoln/newsquadfr'

# exemple 1
ds_newsquad = load_dataset(ds_name)

# exemple 2
data_files = {'train': 'train.json', 'test': 'test.json', 'valid': 'valid.json'}
ds_newsquad = load_dataset(ds_name, data_files=data_files)

# exemple 3
ds_newsquad = load_dataset(ds_name, data_files=data_files, split="valid+test")

(train set)

website	Nb
cnews	20
francetvinfo	40
la-croix	375
lefigaro	160
lemonde	325
lesnumeriques	70
numerama	140
sudouest	475
usinenouvelle	45

Supported Tasks and Leaderboards

• extractive-qa
• open-domain-qa

Languages

Fr-fr

Dataset Structure

Data Instances

{'answers': {'answer_start': [53], 'text': ['manSuvre "agressive']},
 'article_id': 34138,
 'article_title': 'Caricatures, Libye, Haut-Karabakh... Les six dossiers qui '
                  'opposent Emmanuel Macron et Recep Tayyip Erdogan.',
 'article_url': 'https://www.francetvinfo.fr/monde/turquie/caricatures-libye-haut-karabakh-les-six-dossiers-qui-opposent-emmanuel-macron-et-recep-tayyip-erdogan_4155611.html#xtor=RSS-3-[france]',
 'context': 'Dans ce contexte déjà tendu, la France a dénoncé une manSuvre '
            '"agressive" de la part de frégates turques à l\'encontre de l\'un '
            "de ses navires engagés dans une mission de l'Otan, le 10 juin. "
            'Selon Paris, la frégate Le Courbet cherchait à identifier un '
            'cargo suspecté de transporter des armes vers la Libye quand elle '
            'a été illuminée à trois reprises par le radar de conduite de tir '
            "de l'escorte turque.",
 'id': '2261',
 'paragraph_id': 201225,
 'question': "Qu'est ce que la France reproche à la Turquie?",
 'website': 'francetvinfo'}

Data Fields

• answers: a dictionary feature containing:
  • text: a string feature.
  • answer_start: a int64 feature.
• article_id: a int64 feature.
• article_title: a string feature.
• article_url: a string feature.
• context: a string feature.
• id: a string feature.
• paragraph_id: a int64 feature.
• question: a string feature.
• website: a string feature.

Data Splits

Split	Nb
train	1650
test	415
valid	455

Dataset Creation

Curation Rationale

[Needs More Information]

Source Data

Initial Data Collection and Normalization

Paragraphs were chosen according to theses rules:

• parent article must have more than 71% ASCII characters
• paragraphs size must be between 170 and 670 characters
• paragraphs shouldn't contain "A LIRE" or "A VOIR AUSSI"

Then, we stratified our original dataset to create this dataset according to :

• website
• number of named entities
• paragraph size

Who are the source language producers?

[Needs More Information]

Annotations

Annotation process

Using Piaf annotation tools. Three different persons mostly.

Who are the annotators?

Lincoln

Personal and Sensitive Information

[Needs More Information]

Considerations for Using the Data

Social Impact of Dataset

[Needs More Information]

Discussion of Biases

• Annotation is not well controlled
• asking question on news is biaised

Other Known Limitations

[Needs More Information]

Additional Information

Dataset Curators

[Needs More Information]

Licensing Information

https://creativecommons.org/licenses/by-nc-sa/4.0/deed.fr

Citation Information

[Needs More Information]