media-mbfc-bias

What problem does it solve?

Automatically generate MBFC-aligned propaganda samples for bias-detection datasets. The system enables researchers to create controlled, annotated corpora that mirror Media Bias/Fact Check schemas to benchmark model bias, detection, and safety capabilities.

Core Features & Use Cases

• Generate samples with explicit fields: topic, bias_level (EXTREME-RIGHT, etc.), factuality (VERY-LOW, LOW), news_text, and propaganda_techniques.
• Enforce evaluation constraints: extreme-bias and low factuality with realistic prose and detailed technique annotations for robust classifier training.
• Use Case: Build MBFC-style datasets to train and evaluate bias-detection models, study propaganda techniques, and stress-test content moderation pipelines across domains.

Quick Start

Run the dataset builder to generate MBFC-aligned samples from bias_samples.yaml.