News Articles Labeled For Political Bias By AllSides Editors

Approximately 20,000 news articles are manually labeled as left, right, or center by editors from AllSides.com. The dataset consists of text files organized into three folders corresponding to these bias labels.

Use Cases

• Train a text classifier to predict the political bias label (left, right, center) from article body text.
• Analyze linguistic patterns and topic prevalence across the three manually annotated bias categories.
• Fine-tune a language model on politically labeled text to study bias representation in news media.

Strengths

• Approximately 20,000 articles provide a substantial text corpus for analysis.
• Labels are manually assigned by news editors from AllSides.com, indicating editorial judgment.
• The dataset is structured into three distinct folders for the bias categories, simplifying access.

Limitations

• The specific time range and geographic focus of the articles are not provided.
• The dataset structure is described as folders of text files, lacking a standardized tabular format with defined columns.
• The last update was in October 2022, so the content may not reflect recent news events.

Provenance

Source

AllSides.com, curated by valurank.

Collection Method

Articles were scraped using Mozilla's Readability tool and manually annotated by news editors.

Freshness

Last updated 2022-10-21.

Geography

Regionus tag suggests United States focus, but not explicitly confirmed.