Toxicity Detection
Domain: Safety & Ethics
🎯 Motivation
Classifying text as toxic, severe, obscene, threatening, or insulting. Used to filter harmful content.
📋 Concrete Example
Model: ‘This text is 85% toxic, 60% insulting, 30% threatening.’ If toxicity > threshold, flag for review or block. Keeps conversations safe.
🔗 Analogy
A smoke detector - it senses smoke (toxicity) and alarms. Toxicity detection senses harmful language and alerts the system.