What's It Going to Cost You?: Predicting Effort vs. Informativeness for Multi-Label Image Annotations