A privacy-preserving federated learning approach to train robust models for identifying offensive language online without compromising user privacy.