MLM-style-transfer

Bert model is adapted from huggingface https://huggingface.co/transformers/model_doc/bert.html Bert fine-tuning codes are adapted from:https://mccormickml.com/2019/07/22/BERT-fine-tuning/

Data

• Extracted sentences are available https://drive.google.com/drive/folders/1seAMJn3Fh8ZXhDpiW_ih7HDBlqHhIcnu?usp=sharing

To train all mitigation components using default parameters (To define parameters, see src):

• mitigation_model
• generate_neutral_latent_embedding_model
• bias_class_discriminator
  

./train_mitigation_model.sh "binary_classification_data.csv" "bias_only.train.en" "bias_only.dev.en" "neutral_only.train.en" "neutral_only.dev.en"

To mitigate text:

./mitigate_text.sh "bias_only.test.en" "mitigation_model.pth" "generate_neutral_latent_embedding_model.pth" "bias_class_discriminator.pth"

Requirements

torch==1.12.1
transformers==4.21.2
lime==0.2.0.1
sentence_transformers==2.2.2
nltk==3.7
numpy