An Adversarial Approach to Evaluating the Robustness of Event Identification Models