hate speech detection is a common downstream application of natural language
processing (NLP) in the real world. In spite of the increasing accuracy,
current data-driven approaches could easily learn biases from the imbalanced
data distributions originating from humans. The deployment