Modeling disagreement in transformer-based methods for hate speech detection

This GitHub repository contains the work focused on a contrastive learning paradigm to model disagreement in hateful contents. The objective is to create a meaningful representation of the data in order to exploit the embeddings for the final prediction of disagreement.

Run training process

main_supcon.py --batch_size 128 \
 --learning_rate 3e-5 \
 --temp 0.2 \
 --epochs 5

Run final prediction

eval.py