Résumé
In a typical citizen science/crowdsourcing environment, the contributorslabel items. When there are few labels, it is straightforwardto train contributors and judge the quality of their labels bygiving a few examples with known answers. Neither is true whenthere are thousands of domain-specic labels and annotators withheterogeneous skills. This demo paper presents an Active UserTraining framework implemented as a serious game called The-PlantGame. It is based on a set of data-driven algorithms allowingto (i) actively train annotators, and (ii) evaluate the quality of contributors’answers on new test items to optimize predictions.