metadata
task_categories:
- summarization
language:
- fr
size_categories:
- 1K<n<10K
To create this dataset, the test split of MLSUM-FR was filtered by using the code by Aumiller et al. (1) available at https://github.com/dennlinger/summaries/tree/main with following settings
min_length_summary = 18; min_length_reference = 250; length_metric = "whitespace"
The remaining samples were filtered by a function that removed all rows with bi-gram_overlap_fraction between summary and original text >= 0.65, meaning that all summaries in the dataset are on the abstractive side
Finally, 5k random samples were selected and kept in the dataset. All other samples were removed.
(1): Aumiller, D., Fan, J., & Gertz, M. (2023). On the State of German (Abstractive) Text Summarization. arXiv preprint arXiv:2301.07095.