Assessing Fairness in Machine Learning : The Use of Soft Labels to Address Annotator Bias in NLP : A Hate Speech Application

This thesis aims to assess the effects of using soft labels on fairness in machine learning models. The models are trained and evaluated on the task of hate speech detection. In line with this goal, the following main research is addressed during this thesis: "How can a soft label modelling approach, combined with bias detection methods enhance fairness in hate speech detection models?"