README.md

metadata

license: cc-by-nc-sa-4.0
tags:
  - text-classification
  - bias-detection
  - media
  - news
task_categories:
  - text-classification
task_ids:
  - multi-class-classification
language:
  - en
pretty_name: NEWSUNFOLD
size_categories:
  - n<1K

Dataset Card for NEWSUNFOLD

Dataset Summary

NEWSUNFOLD contains sentence-level labels of linguistic media bias collected for the paper
“NewsUnfold: Creating a News-Reading Application That Indicates Linguistic Media Bias and Collects Feedback” (https://doi.org/10.1609/icwsm.v19i1.35847).

The dataset has 310 English news sentences with binary bias labels derived from user feedback on the NewsUnfold web app. Each record includes the sentence text, a bias label, an identifier, the source outlet, the original article URL, and a topic tag.

• Collection window: March 4-11, 2023
• Label meaning: 1 = biased, 0 = not biased
• Selection rule: Only sentences with a clear decision (≥5 votes) are included.

GitHub: https://github.com/Media-Bias-Group/NewsUnfold/

Use Cases

• Training/evaluating bias-detection models
• Auditing or probing linguistic bias signals in news sentences
• Complementing existing datasets (e.g., BABE) with user-feedback–grounded labels

Dataset Structure

Data Fields

• id (string): Unique sentence identifier.
• text (string): The news sentence.
• label (int; 0 or 1): Bias label (1 = biased, 0 = not biased).
• outlet (string): News outlet/source.
• news_link (string/url): Link to the original article.
• topic (string): Topic/category of the article/sentence.

Citation

@article{Hinterreiter_Wessel_Schliski_Echizen_Latoschik_Spinde_2025,
  title    = {NewsUnfold: Creating a News-Reading Application That Indicates Linguistic Media Bias and Collects Feedback},
  volume   = {19},
  url      = {https://doi.org/10.1609/icwsm.v19i1.35847},
  number   = {1},
  journal  = {Proceedings of the International AAAI Conference on Web and Social Media},
  author   = {Hinterreiter, Maximilian and Wessel, Daniel and Schliski, Johanna and Echizen, Isao and Latoschik, Marc Erich and Spinde, Timo},
  year     = {2025},
  doi      = {10.1609/icwsm.v19i1.35847},
  pages    = {1237-1248},
  abstract = {We present NewsUnfold, a news-reading web application that (1) indicates sentence-level linguistic media bias predictions to readers, (2) collects reader feedback on those predictions, and (3) releases a new dataset of sentences with community labels indicating perceived bias. In an in-the-wild study, participants used the app to read curated articles while seeing sentences highlighted as biased or not biased (per a baseline classifier). They could then agree or disagree with these highlights, producing crowd-sourced labels. Using only sentences with a clear decision (≥5 votes), we provide a 310-sentence dataset and report inter-annotator agreement (Krippendorff’s α = .504). We further train a classifier combining the BABE dataset with our new sentences, improving F1 to .824 (+2.49% vs. BABE alone), demonstrating that community feedback can refine bias-detection models and support transparent, user-in-the-loop media analysis.},
  keywords = {media bias, bias detection, user feedback, dataset, news, sentence-level labeling}
}