Awesome075/WCEP-10_parquet

WCEP-10 (Parquet Version)

This repository contains the WCEP-10 dataset in Parquet format. It is a processed version of the dataset originally hosted at ccdv/WCEP-10, which was popularized by the PRIMERA paper.

The dataset is designed for Multi-Document Summarization (MDS) and consists of short, concise summaries derived from the Wikipedia Current Events Portal (WCEP), paired with a cluster of related news articles.

Dataset Summary

• Original Source: WCEP: A Large-Scale Multi-Document Summarization Dataset (Ghalandari et al., 2020)
• Variant Source: PRIMERA: Pyramid-based Masked Sentence Pre-training for Multi-document Summarization (Xiao et al., 2021)
• Format: Parquet
• Language: English

Dataset Structure

The dataset is strictly formatted with two columns, making it ready for immediate use with standard Hugging Face training scripts.

Column	Type	Description
document	string	The input text. Contains the concatenated content of the source articles related to the event.
summary	string	The target summary. A concise, human-written summary of the event from the Wikipedia Current Events Portal.

Data Splits

Split	Number of Instances
Train	8,158
Validation	1,020
Test	1,022

Usage

You can load this dataset directly using the Hugging Face datasets library:

from datasets import load_dataset

dataset = load_dataset("Awesome075/wcep-10_parquet")

print(dataset)
# Output:
# DatasetDict({
#     train: Dataset({
#         features: ['document', 'summary'],
#         num_rows: 8158
#     }),
#     validation: Dataset({
#         features: ['document', 'summary'],
#         num_rows: 1020
#     }),
#     test: Dataset({
#         features: ['document', 'summary'],
#         num_rows: 1022
#     })
# })

Citation

If you use this dataset, please cite both the original WCEP paper and the PRIMERA paper which introduced this specific 10-document subset version.

Original WCEP Dataset:

@inproceedings{WCEP,
    title = "A Large-Scale Multi-Document Summarization Dataset from the Wikipedia Current Events Portal",
    author = "Gholipour Ghalandari, Demian  and
      Hokamp, Chris  and
      Pham, Nghia The  and
      Glover, John  and
      Ifrim, Georgiana",
    booktitle = "Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics",
    month = jul,
    year = "2020",
    address = "Online",
    publisher = "Association for Computational Linguistics",
    url = "https://www.aclweb.org/anthology/2020.acl-main.120",
    pages = "1302--1308",
}

WCEP-10 / PRIMERA:

@inproceedings{PRIMERA,
  title={PRIMERA: Pyramid-based masked sentence pre-training for multi-document summarization},
  author={Xiao, Wen and Beltagy, Iz and Carenini, Giuseppe and Cohan, Arman},
  booktitle={Proceedings of the 60th annual meeting of the association for computational linguistics (Volume 1: Long Papers)},
  pages={5245--5263},
  year={2022}
}