Proposing user-intended adversarial attacks and layer-wise pooling strategies to enhance offensive language detection.
