Made up of news article texts categorized into two distinct subsets for detecting hyperpartisan argumentation. The byarticle collection contains articles with labels derived from crowdsourcing consensus, while the bypublisher collection uses labels based on the reputation of the source as determined by BuzzFeed or MediaBiasFactCheck.com.
Use Cases
- Train a text classification model to distinguish between neutral and hyperpartisan content using the article text and consensus labels
- Compare the performance of bias detection algorithms across the byarticle and bypublisher data splits
- Analyze linguistic patterns of prejudiced allegiance by processing the news article text against the provided bias categories
Strengths
- Includes a byarticle subset where labels are grounded in crowdsourced consensus for specific news texts
- Features a bypublisher subset with labels assigned based on external assessments from BuzzFeed and MediaBiasFactCheck.com
- Specifically targets the identification of hyperpartisan traits such as blind or prejudiced allegiance within news content
- Originally developed as the benchmark for the PAN @ SemEval 2019 Task 4 competition