lunarie16 / DL-Bias

DL-Bias

This is a Deep Learning project about trying out different approaches to test language models, specifically BERT for bias. The main focus is hereby on the sexuality-bias, where is it about finding about how discriminating and harmful language models differ between sexual preferences. The outputs and results reproduce stereotypes and should not be seen as truth.

Code

pip install -r requirements.txt

Appraoches

1. [MASK] Prediction with HONEST evaluation: python bert_honest_eval.py
2. Sentiment Analysis: python sentimentAnalysis.py
3. Contextual Embedding Measure python attributeCheck.py && python calculate_bias_results.py

GitHub Sources/Inspiration

Sentiment Analysis: https://github.com/pysentimiento/pysentimiento/

Contextual Embedding Measure: https://github.com/keitakurita/contextual_embedding_bias_measure