aiml-offensive-benchmark

What problem does it solve?

This ISC template provides a structured benchmark for evaluating offensive-language classification models, ensuring data quality and research reproducibility.

Core Features & Use Cases

• Multi-class coverage: validates the presence of five offense classes (hate_speech, offensive_language, threat_of_violence, dehumanization, incitement) in samples.json.
• Data integrity checks: enforces concrete target_group values, non-placeholder text, and minimum text length to ensure reliable evaluation.
• Automated validation: leverages pytest-based tests (test_benchmark.py) to verify class coverage, text length, target diversity, and offense-specific keywords.

Quick Start

Run the provided pytest suite against samples.json to validate class coverage, text length, target diversity, and offense-specific keywords.