The repository contains code to replicate the experiments in the paper "Robustness and Reliability of Gender Bias Assessment in Word Embeddings: The Role of Base Pairs", by Haiyang Zhang, Alison Sneyd and Mark Stevenson, AACL 2020 (https://www.aclweb.org/anthology/2020.aacl-main.76.pdf).
To run the notebook bias_embeddings_paper_code.ipynb, a copy of the word2vec embeddings GoogleNews-vectors-negative300.bin (available at https://drive.google.com/file/d/0B7XkCwpI5KDYNlNUTTlSS21pQmM/edit) should be placed in the input_data folder. Some of the files in the input_data folder were sourced from https://github.com/tolga-b/debiaswe and http://www.fit.vutbr.cz/~imikolov/rnnlm/word-test.v1.txt.
Requirements:
Python=3.7.6
pandas=1.0.1
numpy=1.18.1
sklearn=0.22.1
gensim=3.8.3
json=2.0.9
matplotlib=3.1.3
re=2.2.1