metadata

license: cc-by-nc-sa-4.0
dataset_info:
  features:
    - name: headline
      dtype: string
    - name: full_text
      dtype: string
    - name: source
      dtype: string
  splits:
    - name: train
      num_bytes: 397400073
      num_examples: 127402
    - name: test
      num_bytes: 33550682
      num_examples: 10000
  download_size: 280794360
  dataset_size: 430950755
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: test
        path: data/test-*
task_categories:
  - text-classification
  - text-generation
  - text-retrieval
  - text2text-generation
  - sentence-similarity
language:
  - it
size_categories:
  - 100K<n<1M

CHANGE-IT

Disclaimer: This dataset is not the official CHANGE-IT repository from EVALITA. For the official dataset and more information, please visit the EVALITA CHANGE-IT page or the CHANGE-IT repository.

Overview

The CHANGE-IT dataset is designed for a style transfer task focused on headlines from Italian newspapers. The dataset comprises approximately 152,000 article-headline pairs sourced from two prominent Italian newspapers, la Repubblica (left-leaning) and Il Giornale (right-leaning). The data is equally split between the two sources, providing a balanced representation of differing political perspectives. For each article, both the headline and its corresponding article text are included.

Purpose

The primary objective of the CHANGE-IT dataset is to facilitate research on style transfer between headlines from newspapers with opposing political orientations. Researchers are encouraged to transform headlines from Il Giornale to the style of la Repubblica and vice versa.

Data Fields

headline: The original headline of the newspaper.
full_text: The article full text associated to the respective headline.
source: The newspaper the sample is coming from, which can be either repubblica or ilgiornale.

Citatation

If you use this dataset, please cite the original authors:

@article{de2020change,
  title={CHANGE-IT@ EVALITA 2020: Change Headlines, Adapt News, GEnerate},
  author={De Mattei, Lorenzo and Cafagna, Michele and AI, Aptus and Dell’Orletta, Felice and Nissim, Malvina and Gatt, Albert},
  year={2020}
}