12,800 human-labeled short statements from politifact.com categorized into six truthfulness levels. Each record pairs a political statement with a professional editor's verdict and a detailed analysis report justifying the label.
Use Cases
- Train a multi-class classifier to categorize statements into six truthfulness levels using the statement text
- Build explainable fact-checking systems that leverage the analysis report feature to provide reasoning for a given label
- Perform linguistic analysis on the statement text to identify markers of deception across different truthfulness labels
Strengths
- 12,800 short statements sourced directly from the politifact.com API
- Six-way truthfulness classification including 'pants-fire' and five other categories ranging from 2,063 to 2,638 instances each
- Detailed analysis reports provided by human editors for every labeled instance to ground judgments