word2vec_gender_professions.py

professions = [
    'director', 'secretary', 'doctor', 'nurse', 'volunteer', 'politician', 'artist', 'scientist',
    'teacher', 'professor', 'parent', 'expert', 'veterinarian', 'physician', 'chef', 'cook',
    'physicist', 'babysitter', 'plumber', 'unemployed',
]

def describe_closer(woman_rank, man_rank):
    if woman_rank == man_rank:
        return "Both are equal!"
    if woman_rank < man_rank:
        return "Woman is {}% closer".format(round((man_rank - woman_rank) * 100 / woman_rank))
    return "Man is {}% closer".format(round((woman_rank - man_rank) * 100 / man_rank))

def print_pretty(profession, woman_rank, man_rank, html=False):
    html_pattern = '<tr>\n' + '<td>{}</td>' * 4 + '\n</tr>'                                  ''
    pattern =  '{:<15} | {:10} | {:10} | {}' if not html else html_pattern

    print(pattern.format(
        profession,
        woman_rank,
        man_rank,
        describe_closer(woman_rank, man_rank)
    ))

from gensim.models import KeyedVectors

# You can download the model here:
# https://github.com/eyaler/word2vec-slim
# The model was trained on 100 billion words from Google News.
print("Loading the model...")
model = KeyedVectors.load_word2vec_format('./GoogleNews-vectors-negative300-SLIM.bin', binary=True)
print("Model loaded.")
print()

print('{:<15} | {:10} | {:10} | {}'.format(
    "profession",
    "woman rank",
    "man rank",
    "comment",
))
print('-' * 60)
for profession in professions:
    woman_rank = model.rank(profession, 'woman')
    man_rank = model.rank(profession, 'man')
    print_pretty(profession, woman_rank, man_rank, html=False)