A Counterfactual Framework for Learning and Evaluating Explanations for Recommender Systems | Synapse