Investigating Gender Bias in Pre-Trained Word Embeddings

This research investigates gender bias in word embeddings, using the GloVe 6B dataset created from Wikipedia and Gigaword Corpus. Through converting GloVe to Word2Vec and performing bias tests via analogy and similarity metrics, the study applies Python and Gensim for analysis, focusing on professions. Evaluation includes semantic sanity checks and bias testing, revealing biases in embedding representations. Outcomes highlight ethical considerations in NLP, and future work suggests broader...