MaLA-LM/MassiveSumm_long

MassiveSumm Long Subset

This dataset is a subset of MassiveSumm by subsampling and setting a maximum sequence length of 7.5k tokens. Links to reproduce the whole set of MassiveSumm via Common Crawl and the Wayback Machine are provided in the repository of MassiveSumm.

Description: MassiveSumm is a very large-scale, highly multilingual news summarization dataset designed to support summarization research across a wide range of languages. It encompasses 92 diverse languages and 35 writing scripts, aiming to provide a diverse data foundation for multilingual summarization.

Original Data: Redistributing data from web is a tricky matter. The authors of the original data are working on providing efficient access to the entire dataset, as well as expanding it even further. For the time being they only provide links to reproduce subsets of the entire dataset through either common crawl and the wayback machine. The dataset is also available upon request ([email protected]). The redistribution of this subset, MassiveSumm_long, aims at producing a practical setting for LLM evaluation.

Data Fields:

• url: The URL of the original news article.
• title: The title of the news article.
• text: The original news article text in the respective language.
• summary: The extracted or generated summary of the news article in the respective language.
• language: The language of the article and summary.
• date: The date when the article was crawled.

Caveats: The data is noted to be noisy and recall-oriented. The creators recommend reading their analysis on the efficacy of this type of data collection method before use.

Intended Use: Research in multilingual news summarization, cross-lingual transfer learning for summarization, and evaluation of multilingual models.

Take-Down Policy:

The dataset maintainers are committed to addressing legitimate concerns regarding the dataset's content. We will comply to legitimate requests by removing the affected texts.

Citation:

@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",
}