---
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: validation
    path: data/validation-*
  - split: test
    path: data/test-*
task_categories:
- summarization
language:
- fr
size_categories:
- 100K<n<1M
license: cc-by-4.0
dataset_info:
  features:
  - name: text
    dtype: string
  - name: summary
    dtype: string
  - name: dataset
    dtype: string
  splits:
  - name: train
    num_bytes: 3463599993
    num_examples: 788358
  - name: validation
    num_bytes: 112076027
    num_examples: 26935
  - name: test
    num_bytes: 118025713
    num_examples: 26380
  download_size: 2250524647
  dataset_size: 3693701733
---

# Dataset information 
**Dataset concatenating Summarization datasets available in French and open-source.**  
There are a total of **841,673** rows, of which 788,358 are for training, 26,935 for validation and 26,380 for testing.  


#  Usage
```
from datasets import load_dataset
dataset = load_dataset("CATIE-AQ/frenchSUM")
```


# Dataset
## Details of rows
| Dataset Original | Splits | Note  |
| ----------- | ----------- | ----------- |
| [GEM/xwikis](https://huggingface.co/datasets/GEM/xwikis)| 323,726 train / 7,339 validation / 6,992 test | We only keep the French split  |           
| [maurya/alpaca_motivations_sommaires](https://huggingface.co/datasets/maurya/alpaca_motivations_sommaires)| 7,801 train | |       
| [csebuetnlp/CrossSum](https://huggingface.co/datasets/csebuetnlp/CrossSum)| 8,648 train / 1,083 validation / 1,083 test | We only keep the French split<br>Perhaps not a great dataset<br>(there's information in the summary that isn't in the original text)  |       
| [csebuetnlp/xlsum](https://huggingface.co/datasets/csebuetnlp/xlsum)| 8,697 train / 1,086 validation / 1,086 test | We only keep the French split |       
| [orange_sum](https://huggingface.co/datasets/orange_sum)| 21,401 train / 1,500 validation / 1,500 test | We use the `abstract` split |         
| [mlsum](https://huggingface.co/datasets/mlsum)| 392,017 train / 16,059 validation / 15,820 test | We only keep the French split |       
| [fr_covid_news_8k](https://www.kaggle.com/datasets/guscor/covid19-french-news-dataset)| 8,084 train |  |       
| [FakeNewsDetectionFr](https://github.com/jeugregg/FakeNewsDetectionFr)| 1,260 train |  |       
| [MassiveSumm_short](https://huggingface.co/datasets/MaLA-LM/MassiveSumm_short)| 2,500 train | We only keep the French split  |      
| [MassiveSumm_long](https://huggingface.co/datasets/MaLA-LM/MassiveSumm_long)| 684 train | We only keep the French split   |      
| [wikinews](https://huggingface.co/datasets/malteos/wikinews)| 22,463 train | We only keep the French split   |      


## Removing duplicate data and leaks
The sum of the values of the datasets listed here gives the following result:

```
DatasetDict({
    train: Dataset({
        features: ['text', 'summary', 'dataset'],
        num_rows: 797281
    })
    validation: Dataset({
        features: ['text', 'summary', 'dataset'],
        num_rows: 27067
    })
    test: Dataset({
        features: ['text', 'summary', 'dataset'],
        num_rows: 26481
    })
})
```

However, a data item in training split A may not be in A's test split, but may be present in B's test set, creating a leak when we create the A+B dataset.  
The same logic applies to duplicate data. So we need to make sure we remove them.  
After our clean-up, we finally have the following numbers:

```
DatasetDict({
    train: Dataset({
        features: ['text', 'summary', 'dataset'],
        num_rows: 788358
    })
    validation: Dataset({
        features: ['text', 'summary', 'dataset'],
        num_rows: 26935
    })
    test: Dataset({
        features: ['text', 'summary', 'dataset'],
        num_rows: 26380
    })
})
```

Note that an additional cleanup has been applied to all rows, removing all URLs and emojis present in the articles.

## Columns
```
dataset_train = dataset['train'].to_pandas()
dataset_train.head()

    text                                                   summary                                             dataset
0 	L'astronomie est considérée comme la plus anci... 	L’astronomie est la science de l’observation d...   xwikis
1 	L'origine du mot est anglaise, "tourist", qui ... 	Le mot tourisme désigne le fait de voyager pou...   xwikis
2 	Étymologiquement, le mot français "philosophie... 	La philosophie, du grec ancien "φιλοσοφία" (com...   xwikis
3 	Le terme biologie est formé par la composition... 	La biologie (du grec "bios" « la vie » et "log...   xwikis
4 	Trois étymologies sont fréquemment citées, mai... 	La chimie est une science de la nature qui étu...   xwikis
```

- the `text` column contains the text
- the `summary` column contains text summary
- the `dataset` column identifies the row's original dataset (if you wish to apply filters to it)

## Split
- `train` corresponds to the concatenation of `xwikis` + `alpaca_motivations_sommaires` + `cross_sum` + `xlsum` + `orange_sum` + `mlsum` + `fr_covid_news_8k` + `FakeNewsDetectionFr` + `MassiveSumm_short`+ `MassiveSumm_long` + `wikinews`
- `validation` corresponds to the concatenation of `xwikis` + `cross_sum` + `xlsum` + `orange_sum` + `mlsum`
- `test` corresponds to the concatenation of `xwikis` + `cross_sum` + `xlsum` + `orange_sum` + `mlsum`



# Citations

### alpaca_motivations_sommaires
```
Dataset by Amaury FOURET (2023) 
Hugging Face repository: https://huggingface.co/datasets/maurya/alpaca_motivations_sommaires
```

### FakeNewsDetectionFr
```
Dataset by Gregory LANG (2019) 
Hugging Face repository: https://github.com/jeugregg/FakeNewsDetectionFr
```

### fr_covid_news_8k
```
@dataset{covidfrenchnews,  
  author = {Gustave Cortal},  
  year = {2021},  
  month = {03},  
  title = {COVID-19: French news dataset},  
  url = {https://www.gustavecortal.com}}
```

### CrossSum
```
@misc{bhattacharjee2023crosssum,  
      title={CrossSum: Beyond English-Centric Cross-Lingual Summarization for 1,500+ Language Pairs},  
      author={Abhik Bhattacharjee and Tahmid Hasan and Wasi Uddin Ahmad and Yuan-Fang Li and Yong-Bin Kang and Rifat Shahriyar},  
      year={2023},  
      eprint={2112.08804},  
      archivePrefix={arXiv},  
      primaryClass={cs.CL}}
```

### mlsum
```
@inproceedings{scialom-etal-2020-mlsum,  
    title = "{MLSUM}: The Multilingual Summarization Corpus",   
    author = "Scialom, Thomas and Dray, Paul-Alexis  and Lamprier, Sylvain and Piwowarski, Benjamin and Staiano, Jacopo",
    editor = "Webber, Bonnie and Cohn, Trevor and He, Yulan and Liu, Yang",  
    booktitle = "Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP)",  
    month = nov,  
    year = "2020",  
    address = "Online",  
    publisher = "Association for Computational Linguistics",  
    url = "https://aclanthology.org/2020.emnlp-main.647",  
    doi = "10.18653/v1/2020.emnlp-main.647",  
    pages = "8051--8067"}
```

### orange_sum
```
@article{eddine2020barthez,  
  title={BARThez: a Skilled Pretrained French Sequence-to-Sequence Model},  
  author={Eddine, Moussa Kamal and Tixier, Antoine J-P and Vazirgiannis, Michalis},  
  journal={arXiv preprint arXiv:2010.12321},  
  year={2020}}
```

### xlsum
```
@inproceedings{hasan-etal-2021-xl,  
    title = "{XL}-Sum: Large-Scale Multilingual Abstractive Summarization for 44 Languages",  
    author = "Hasan, Tahmid and Bhattacharjee, Abhik and Islam, Md. Saiful  and Mubasshir, Kazi and Li, Yuan-Fang and Kang, Yong-Bin and Rahman, M. Sohel and Shahriyar, Rifat",  
    editor = "Zong, Chengqing  and Xia, Fei and Li, Wenjie and Navigli, Roberto",  
    booktitle = "Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021",  
    month = aug,  
    year = "2021",  
    address = "Online",  
    publisher = "Association for Computational Linguistics",  
    url = "https://aclanthology.org/2021.findings-acl.413",  
    doi = "10.18653/v1/2021.findings-acl.413",  
    pages = "4693--4703"}
```

### xwikis
```
@misc{perezbeltrachini2022models,  
      title={Models and Datasets for Cross-Lingual Summarisation},  
      author={Laura Perez-Beltrachini and Mirella Lapata},  
      year={2022},  
      eprint={2202.09583},  
      archivePrefix={arXiv},  
      primaryClass={cs.CL}}
```

### MassiveSumm
```
@inproceedings{djamshidi-etal-2021-massivesumm,
title = "MassiveSumm: A Very Large-Scale, Very Multilingual News Summarisation Dataset",
author = "Djamshidi, Amir Abbas and Mehrabi, Pegah and Paun, Andrei and Sagot, Beno{^i}t",
booktitle = "Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing",
month = nov,
year = "2021",
address = "Online and Punta Cana, Dominican Republic",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2021.emnlp-main.303",
pages = "3749--3765",
}
```

### wikinews
```
Dataset by Malte OSTENDORFF (2024) 
Hugging Face repository: https://huggingface.co/datasets/malteos/wikinews
```

### FrenchSUM
```
@misc {frenchSUM,   
    author       = { {BOURDOIS, Loïck} },  
    organization  = { {Centre Aquitain des Technologies de l'Information et Electroniques} },  
    title        = { frenchSUM (Revision 299b07c) },  
    year         = 2023,  
    url          = { https://huggingface.co/datasets/CATIE-AQ/frenchSUM },  
    doi          = { 10.57967/hf/6447 },  
    publisher    = { Hugging Face } }
```

# License
[cc-by-4.0](https://creativecommons.org/licenses/by/4.0/deed.en)