The articles crawled from www.allsides.com are available in the ./data folder, along with the different evaluation splits.

The dataset consists of a total of 37,554 articles. Each article is stored as a JSON object in the ./data/jsons directory, and contains the following fields:

1. ID: an alphanumeric identifier.
2. topic: the topic being discussed in the article.
3. source: the name of the articles's source (example: New York Times)
4. source_url: the URL to the source's homepage (example: www.nytimes.com)
5. url: the link to the actual article.
6. date: the publication date of the article.
7. authors: a comma-separated list of the article's authors.
8. title: the article's title.
9. content_original: the original body of the article, as returned by the newspaper3k Python library.
10. content: the processed and tokenized content, which is used as input to the different models.
11. bias_text: the label of the political bias annotation of the article (left, center, or right).
12. bias: the numeric encoding of the political bias of the article (0, 1, or 2).

The ./data/splits directory contains the two types of splits, as discussed in the paper: random and media-based. For each of these types, we provide the train, validation and test files that contains the articles' IDs belonging to each set, along with their numeric bias label.

Under maintenance. To be available soon.

@inproceedings{baly2020we,
  author      = {Baly, Ramy and Da San Martino, Giovanni and Glass, James and Nakov, Preslav},
  title       = {We Can Detect Your Bias: Predicting the Political Ideology of News Articles},
  booktitle   = {Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP)},
  series      = {EMNLP~'20},
  NOmonth     = {November},
  year        = {2020}
  pages       = {4982--4991},
  NOpublisher = {Association for Computational Linguistics}
}