Explanations as mechanisms for supporting algorithmic transparency