Debiasing Framework for Vector Representations

Contributer: Peini Cheng

This repository is an implementation and evaluation of an existing work published at NAACL 2024 https://aclanthology.org/2024.naacl-long.332.pdf

1. Download datasets

The datasets required for this project include the original GloVe vector representations and the debiased INLP vector representations.

PLease make sure you have at least 3GB of free space in the download path.

Please download the pre-processed datasets from my Google drive shared files https://drive.google.com/file/d/1CHBOGgibMgL2R27B3MI1yBiua36OfBw-/view?usp=sharing. After downloading the vector_representations zip file, extract the two text files glove.6B.300d.txt and inlp.6B.300d.txt out of the folder and place them under the root folder of this project.

debiasing_representation
├── README.md
├── requirement.txt
├── main.py
├── t-sne.jpg
├── glove.6B.300d.txt
├── inlp.6B.300d.txt

Please let me know if there's any problem with downloading the datasets.

2. Check requirements

This project uses Python 3.12.6

Library	Version
numpy	2.2.4
tqdm	4.67.1
matplotlib	3.10.1

(This project does not require any machine learning libraries) To install these libraries in Python, run pip install -r requirement.txt

3. Run the code

Make sure you are in the debiasing_representation folder, and run python main.py on Windows or python3 main.py on other platforms. It may take a few minutes for the results to show, depending on the PC setup.