The process of choosing which training corpora or datasets to use for a model, typically to maximize performance on a target task.