Abstract

Gold HITs — Human Intelligence Tasks with known answers — are commonly used to measure worker performance and data quality in industrial applications of crowdsourcing. We suggest adaptive exploration as a promising approach for automated, scalable Gold HIT creation. We substantiate this with initial experiments in a stylized model.