Datasets:
language:
- fr
license: cc-by-nc-sa-4.0
Dataset origin: https://github.com/adrianchifu/FreSaDa
FreSaDa: The French Satire Data Set
The FreSaDa data set contains regular and satirical samples of text collected from the French news domain.
Description
General Information
FreSaDa, the French Satire Data Set, is composed of 11,570 articles from the newsdomain. The news articles are of two types: satirical and regular. Two possible tasks may be considered on FreSaDa:
- Cross-domain binary classification of full news articles into regular versus satirical examples
- Cross-domain binary classification of headlines into regular versus satirical examples
The data set is divided into three subsets:
- training (8,716 samples)
- testing (2,854 samples)
Each sample contains the news article's title and text, as well as the corresponding label.
Data Organization
The data set is divided in two folders, train and test, corresponding to the two subsets for training and testing. In each folder there is a subfolder entitled texts, containing the texts of the news articles. Each folder also contains a file called summary.tsv:
The labels are associated as follows:
- 1 => Satiric News
- -1 => Regular News
If the experiments require a validation subset, the test subset may be divided into two equal parts:
- the samples with odd rank (1st, 3rd, 5th, ...) for validation (1,427 samples)
- the samples with even rank (2nd, 4th, 6th, ...) for testing (1,427 samples)
Citation
[1] Radu Tudor Ionescu, Adrian Gabriel Chifu. FreSaDa: A French Satire Data Set for Cross-Domain Satire Detection. In: The International Joint Conference on Neural Network, IJCNN 2021 (2021). (link to article)
BibTeX citation:
@inproceedings{IonescuChifu2021IJCNN,
author = {Ionescu, Radu-Tudor and Chifu, Adrian-Gabriel},
title = {FreSaDa: A French Satire Data Set for Cross-Domain Satire Detection},
year = {2021},
booktitle = {The International Joint Conference on Neural Network, IJCNN 2021},
series = {IJCNN2021}
}