Skip to content
Permalink
Branch: master
Find file Copy path
Find file Copy path
Fetching contributors…
Cannot retrieve contributors at this time
574 lines (464 sloc) 22.6 KB
import math
from collections import Counter
import nltk
from scipy.stats import chi2
from gender_novels.common import store_pickle, load_pickle
from gender_novels.corpus import Corpus
# TODO: Rewrite all of this using a Dunning class in a non-messy way.
def dunn_individual_word(total_words_in_corpus_1, total_words_in_corpus_2,
count_of_word_in_corpus_1,
count_of_word_in_corpus_2):
'''
applies dunning log likelihood to compare individual word in two counter objects
:param word: desired word to compare
:param m_corpus: c.filter_by_gender('male')
:param f_corpus: c. filter_by_gender('female')
:return: log likelihoods and p value
>>> total_words_m_corpus = 8648489
>>> total_words_f_corpus = 8700765
>>> wordcount_female = 1000
>>> wordcount_male = 50
>>> dunn_individual_word(total_words_m_corpus,total_words_f_corpus,wordcount_male,wordcount_female)
-1047.8610274053995
'''
a = count_of_word_in_corpus_1
b = count_of_word_in_corpus_2
c = total_words_in_corpus_1
d = total_words_in_corpus_2
e1 = c * (a + b) / (c + d)
e2 = d * (a + b) / (c + d)
dunning_log_likelihood = 2 * (a * math.log(a / e1) + b * math.log(b / e2))
if count_of_word_in_corpus_1 * math.log(count_of_word_in_corpus_1 / e1) < 0:
dunning_log_likelihood = -dunning_log_likelihood
p = 1 - chi2.cdf(abs(dunning_log_likelihood),1)
return dunning_log_likelihood
def dunning_total(counter1, counter2, filename_to_pickle=None):
'''
runs dunning_individual on words shared by both counter objects
(-) end of spectrum is words for counter_2
(+) end of spectrum is words for counter_1
the larger the magnitude of the number, the more distinctive that word is in its
respective counter object
use filename_to_pickle to store the result so it only has to be calculated once and can be
used for multiple analyses.
>>> from collections import Counter
>>> female_counter = Counter({'he': 1, 'she': 10, 'and': 10})
>>> male_counter = Counter({'he': 10, 'she': 1, 'and': 10})
>>> results = dunning_total(female_counter, male_counter)
# Results is a dict that maps from terms to results
# Each result dict contains the dunning score...
>>> results['he']['dunning']
-8.547243830635558
# ... counts for corpora 1 and 2 as well as total count
>>> results['he']['count_total'], results['he']['count_corp1'], results['he']['count_corp2']
(11, 1, 10)
# ... and the same for frequencies
>>> results['he']['freq_total'], results['he']['freq_corp1'], results['he']['freq_corp2']
(0.2619047619047619, 0.047619047619047616, 0.47619047619047616)
:return: dict
'''
total_words_counter1 = 0
total_words_counter2 = 0
#get word total in respective counters
for word1 in counter1:
total_words_counter1 += counter1[word1]
for word2 in counter2:
total_words_counter2 += counter2[word2]
#dictionary where results will be returned
dunning_result = {}
for word in counter1:
counter1_wordcount = counter1[word]
if word in counter2:
counter2_wordcount = counter2[word]
if counter1_wordcount + counter2_wordcount < 10:
continue
dunning_word = dunn_individual_word( total_words_counter1, total_words_counter2,
counter1_wordcount,counter2_wordcount)
dunning_result[word] = {
'dunning': dunning_word,
'count_total': counter1_wordcount + counter2_wordcount,
'count_corp1': counter1_wordcount,
'count_corp2': counter2_wordcount,
'freq_total': (counter1_wordcount + counter2_wordcount) / (total_words_counter1 +
total_words_counter2),
'freq_corp1': counter1_wordcount / total_words_counter1,
'freq_corp2': counter2_wordcount / total_words_counter2
}
if filename_to_pickle:
store_pickle(dunning_result, filename_to_pickle)
return dunning_result
def male_vs_female_authors_analysis_dunning_lesser():
'''
tests word distinctiveness of shared words between male and female corpora using dunning
:return: dictionary of common shared words and their distinctiveness
'''
c = Corpus('test_corpus')
m_corpus = c.filter_by_gender('male')
f_corpus = c.filter_by_gender('female')
wordcounter_male = m_corpus.get_wordcount_counter()
wordcounter_female = f_corpus.get_wordcount_counter()
results = dunning_total(wordcounter_male, wordcounter_female)
print("women's top 10: ", results[0:10])
print("men's top 10: ", list(reversed(results[-10:])))
return results
def dunning_result_displayer(dunning_result, number_of_terms_to_display=10,
corpus1_display_name=None, corpus2_display_name=None,
part_of_speech_to_include=None):
"""
Convenience function to display dunning results as tables.
part_of_speech_to_include can either be a list of POS tags or a 'adjectives, 'adverbs',
'verbs', or 'pronouns'. If it is None, all terms are included.
:param dunning_result: Dunning result dict to display
:param number_of_terms_to_display: Number of terms for each corpus to display
:param corpus1_display_name: Name of corpus 1 (e.g. "Female Authors")
:param corpus2_display_name: Name of corpus 2 (e.g. "Male Authors")
:param part_of_speech_to_include: e.g. 'adjectives', or 'verbs'
:return:
"""
pos_names_to_tags = {
'adjectives': ['JJ', 'JJR', 'JJS'],
'adverbs': ['RB', 'RBR', 'RBS', 'WRB'],
'verbs': ['VB', 'VBD', 'VBG', 'VBN', 'VBP', 'VBZ'],
'pronouns': ['PRP', 'PRP$', 'WP', 'WP$']
}
if part_of_speech_to_include in pos_names_to_tags:
part_of_speech_to_include = pos_names_to_tags[part_of_speech_to_include]
if not corpus1_display_name:
corpus1_display_name = 'Corpus 1'
if not corpus2_display_name:
corpus2_display_name = 'Corpus 2'
headings = ['term', 'dunning', 'count_total', 'count_corp1', 'count_corp2', 'freq_total',
'freq_corp1', 'freq_corp2']
output = f'\nDisplaying Part of Speech: {part_of_speech_to_include}\n'
for i, corpus_name in enumerate([corpus1_display_name, corpus2_display_name]):
output += f'\nDunning Log-Likelihood results for {corpus_name}\n|'
for heading in headings:
heading = heading.replace('_corp1', ' ' + corpus1_display_name).replace('_corp2',
' ' + corpus2_display_name)
output += ' {:19s}|'.format(heading)
output += '\n' + 8 * 21 * '_' + '\n'
reverse = True
if i == 1: reverse = False
sorted_results = sorted(dunning_result.items(), key=lambda x: x[1]['dunning'],
reverse=reverse)
count_displayed = 0
for result in sorted_results:
if count_displayed == number_of_terms_to_display:
break
term = result[0]
term_pos = nltk.pos_tag([term])[0][1]
if part_of_speech_to_include and term_pos not in part_of_speech_to_include:
continue
output += '| {:18s}|'.format(result[0])
for heading in headings[1:]:
if heading in ['freq_total', 'freq_corp1', 'freq_corp2']:
output += ' {:16.4f}% |'.format(result[1][heading] * 100)
elif heading in ['dunning']:
output += ' {:17.2f} |'.format(result[1][heading])
else:
output += ' {:17.0f} |'.format(result[1][heading])
output += '\n'
count_displayed += 1
print(output)
def compare_word_association_in_corpus_analysis_dunning(word1, word2, corpus=None,
corpus_name=None):
"""
Uses Dunning analysis to compare words associated with word1 vs words associated with word2 in
the Corpus passed in as the parameter. If a corpus and corpus_name are passsed in, then the
analysis will use the corpus but name the file after corpus_name. If no corpus is passed in but
a corpus_name is, then the method will try to create a Corpus by corpus = Corpus(corpus_name).
If neither a corpus nor a corpus_name is passed in, analysis is simply done on the Gutenberg
corpus.
:param word1: str
:param word2: str
:param corpus: Corpus
:param corpus_name: str
:return: dict
"""
if corpus:
if not corpus_name:
corpus_name = corpus.corpus_name
else:
if not corpus_name:
corpus_name = "gutenberg"
corpus = Corpus(corpus_name)
pickle_filename = f'dunning_{word1}_vs_{word2}_associated_words_{corpus_name}'
try:
results = load_pickle(pickle_filename)
except IOError:
try:
pickle_filename = f'dunning_{word2}_vs_{word1}_associated_words_{corpus_name}'
results = load_pickle(pickle_filename)
except:
word1_counter = Counter()
word2_counter = Counter()
for novel in corpus.novels:
word1_counter.update(novel.words_associated(word1))
word2_counter.update(novel.words_associated(word2))
results = dunning_total(word1_counter, word2_counter,
filename_to_pickle=pickle_filename)
for group in [None, 'verbs', 'adjectives', 'pronouns', 'adverbs']:
dunning_result_displayer(results, number_of_terms_to_display=50,
part_of_speech_to_include=group)
return results
def compare_word_association_between_corpus_analysis_dunning(word, corpus1=None, corpus1_name=None,
corpus2=None, corpus2_name=None, use_word_window=False, word_window=None):
"""
Uses Dunning analysis to compare words associated with word between corpuses. If a corpus and corpus_name are
passsed in, then the analysis will use the corpus but name the file after corpus_name. If no corpus is passed in but
a corpus_name is, then the method will try to create a Corpus by corpus = Corpus(corpus_name).
If neither a corpus nor a corpus_name is passed in, analysis is simply done on the Gutenberg
corpus.
:param word1: str
:param corpus: Corpus
:param corpus_name: str
:return: dict
"""
if corpus1:
if not corpus1_name:
corpus1_name = corpus1.corpus_name
else:
if not corpus1_name:
corpus1_name = "gutenberg"
corpus1 = Corpus(corpus1_name)
if corpus2:
if not corpus2_name:
corpus2_name = corpus2.corpus_name
else:
if not corpus2_name:
corpus2_name = "gutenberg"
corpus2 = Corpus(corpus2_name)
pickle_filename = (f'dunning_{word}_associated_words_{corpus1_name}_vs_{corpus2_name}_in_'
f'{corpus1.corpus_name}')
if use_word_window:
pickle_filename+= f'_word_window_{word_window}'
try:
results = load_pickle(pickle_filename)
except IOError:
print("Precalculated result not available. Running analysis now...")
corpus1_counter = Counter()
corpus2_counter = Counter()
for novel in corpus1.novels:
if use_word_window:
get_word_windows(self, search_terms, window_size=word_window)
else:
corpus1_counter.update(novel.words_associated(word))
for novel in corpus2.novels:
if use_word_window:
get_word_windows(self, search_terms, window_size=word_window)
else:
corpus2_counter.update(novel.words_associated(word))
results = dunning_total(corpus1_counter, corpus2_counter,
filename_to_pickle=pickle_filename)
for group in [None, 'verbs', 'adjectives', 'pronouns', 'adverbs']:
dunning_result_displayer(results, number_of_terms_to_display=20,
corpus1_display_name=f'{corpus1_name}. {word}',
corpus2_display_name=f'{corpus2_name}. {word}',
part_of_speech_to_include=group)
return results
def male_VS_female_analysis_dunning(corpus_name, display_data = False):
'''
tests word distinctiveness of shared words between male and female corpora using dunning
Prints out the most distinctive terms overall as well as grouped by verbs, adjectives etc.
:return: dict
'''
# By default, try to load precomputed results. Only calculate if no stored results are
# available.
pickle_filename = f'dunning_male_vs_female_chars_{corpus_name}'
try:
results = load_pickle(pickle_filename)
except IOError:
c = Corpus(corpus_name)
m_corpus = c.filter_by_gender('male')
f_corpus = c.filter_by_gender('female')
from collections import Counter
wordcounter_male = Counter()
wordcounter_female = Counter()
for novel in m_corpus:
wordcounter_male += novel.words_associated('he')
for novel in f_corpus:
wordcounter_female += novel.words_associated('he')
# wordcounter_male = m_corpus.get_wordcount_counter()
# wordcounter_female = f_corpus.get_wordcount_counter()
results = dunning_total(wordcounter_male, wordcounter_female,
filename_to_pickle=pickle_filename)
if display_data:
for group in [None, 'verbs', 'adjectives', 'pronouns', 'adverbs']:
dunning_result_displayer(results, number_of_terms_to_display=20,
corpus1_display_name='Fem Author',
corpus2_display_name='Male Author',
part_of_speech_to_include=group)
return results
def dunning_result_to_dict(dunning_result, number_of_terms_to_display=10,
part_of_speech_to_include=None):
'''
Receives a dictionary of results and returns a dictionary of the top
number_of_terms_to_display most distinctive results for each corpus that have a part of speech
matching part_of_speech_to_include
:param dunning_result: Dunning result dict that will be sorted through
:param number_of_terms_to_display: Number of terms for each corpus to display
:param part_of_speech_to_include: e.g. 'adjectives', or 'verbs'
:return: dict
'''
pos_names_to_tags = {
'adjectives': ['JJ', 'JJR', 'JJS'],
'adverbs': ['RB', 'RBR', 'RBS', 'WRB'],
'verbs': ['VB', 'VBD', 'VBG', 'VBN', 'VBP', 'VBZ'],
'pronouns': ['PRP', 'PRP$', 'WP', 'WP$']
}
if part_of_speech_to_include in pos_names_to_tags:
part_of_speech_to_include = pos_names_to_tags[part_of_speech_to_include]
final_results_dict = {}
reverse = True
for i in range(2):
sorted_results = sorted(dunning_result.items(), key=lambda x: x[1]['dunning'],
reverse=reverse)
count_displayed = 0
for result in sorted_results:
if count_displayed == number_of_terms_to_display:
break
term = result[0]
term_pos = nltk.pos_tag([term])[0][1]
if part_of_speech_to_include and term_pos not in part_of_speech_to_include:
continue
final_results_dict[result[0]]=result[1]
count_displayed += 1
reverse = False
return final_results_dict
################################################
# Individual Analyses #
################################################
# Male Authors versus Female Authors
################################################
def male_vs_female_authors_analysis_dunning(corpus_name, display_results=False):
'''
tests word distinctiveness of shared words between male and female authors using dunning
If called with display_results=True, prints out the most distinctive terms overall as well as
grouped by verbs, adjectives etc.
Returns a dict of all terms in the corpus mapped to the dunning data for each term
:return:dict
'''
# By default, try to load precomputed results. Only calculate if no stored results are
# available.
pickle_filename = f'dunning_male_vs_female_authors_{corpus_name}'
try:
results = load_pickle(pickle_filename)
except IOError:
c = Corpus(corpus_name)
m_corpus = c.filter_by_gender('male')
f_corpus = c.filter_by_gender('female')
wordcounter_male = m_corpus.get_wordcount_counter()
wordcounter_female = f_corpus.get_wordcount_counter()
results = dunning_total(wordcounter_female, wordcounter_male,
filename_to_pickle=pickle_filename)
if display_results:
for group in [None, 'verbs', 'adjectives', 'pronouns', 'adverbs']:
dunning_result_displayer(results, number_of_terms_to_display=20,
corpus1_display_name='Fem Author',
corpus2_display_name='Male Author',
part_of_speech_to_include=group)
return results
# Male Characters versus Female Characters (words following 'he' versus words following 'she')
##############################################################################################
def he_vs_she_associations_analysis_dunning(corpus_name):
"""
Uses Dunning analysis to compare words associated with 'he' vs words associated with 'she' in
the Corpus passed in as the parameter. The corpus_name parameter is if you want to name the file
something other than Gutenberg (e.g. Gutenberg_female_authors)
:param corpus_name: str
"""
corpus = Corpus(corpus_name)
pickle_filename = f'dunning_he_vs_she_associated_words_{corpus_name}'
try:
results = load_pickle(pickle_filename)
except IOError:
he_counter = Counter()
she_counter = Counter()
for novel in corpus.novels:
he_counter.update(novel.words_associated("he"))
she_counter.update(novel.words_associated("she"))
results = dunning_total(she_counter, he_counter, filename_to_pickle=pickle_filename)
for group in [None, 'verbs', 'adjectives', 'pronouns', 'adverbs']:
dunning_result_displayer(results, number_of_terms_to_display=20,
corpus1_display_name='she...',
corpus2_display_name='he..',
part_of_speech_to_include=group)
# Female characters as written by Male Authors versus Female Authors
####################################################################
def female_characters_author_gender_differences(corpus_name):
"""
Compares how male authors versus female authors write female characters by looking at the words
that follow 'she'
:param corpus_name:
:return:
"""
male_corpus = Corpus(corpus_name).filter_by_gender('male')
female_corpus = Corpus(corpus_name).filter_by_gender('female')
compare_word_association_between_corpus_analysis_dunning(word='she',
corpus1=female_corpus, corpus1_name='fem aut',
corpus2=male_corpus, corpus2_name='male aut')
# Male characters as written by Male Authors versus Female Authors
####################################################################
def male_characters_author_gender_differences(corpus_name):
"""
Compares how male authors versus female authors write male characters by looking at the words
that follow 'he'
:param corpus_name:
:return:
"""
male_corpus = Corpus(corpus_name).filter_by_gender('male')
female_corpus = Corpus(corpus_name).filter_by_gender('female')
compare_word_association_between_corpus_analysis_dunning(word='he',
corpus1=female_corpus, corpus1_name='female aut',
corpus2=male_corpus, corpus2_name='male aut')
# God as written by Male Authors versus Female Authors
####################################################################
def god_author_gender_differences(corpus_name):
"""
Compares how male authors versus female authors refer to God by looking at the words
that follow 'God'
:param corpus_name:
:return:
"""
male_corpus = Corpus(corpus_name).filter_by_gender('male')
female_corpus = Corpus(corpus_name).filter_by_gender('female')
compare_word_association_between_corpus_analysis_dunning(word='God',
corpus1=female_corpus, corpus1_name='female aut',
corpus2=male_corpus, corpus2_name='male aut')
def money_author_gender_differences(corpus_name):
"""
Compares how male authors versus female authors refer to money by looking at the words
before and after money'
:param corpus_name:
:return:
"""
male_corpus = Corpus(corpus_name).filter_by_gender('male')
female_corpus = Corpus(corpus_name).filter_by_gender('female')
compare_word_association_between_corpus_analysis_dunning(word=['money','dollars', 'pounds', 'euros', 'dollar', 'pound','euro', 'wealth', 'income'],
corpus1=female_corpus, corpus1_name='female aut',
corpus2=male_corpus, corpus2_name='male aut')
# America as written by Male Authors versus Female Authors
####################################################################
def america_author_gender_differences(corpus_name):
"""
Compares how American male authors versus female authors refer to America by looking at the words
that follow 'America'
:param corpus_name:
:return:
"""
male_corpus = Corpus(corpus_name).filter_by_gender('male')
female_corpus = Corpus(corpus_name).filter_by_gender('female')
compare_word_association_between_corpus_analysis_dunning(word='America',
corpus1=female_corpus, corpus1_name='female aut',
corpus2=male_corpus, corpus2_name='male aut')
if __name__ == '__main__':
#### Uncomment any of the lines below to run one of the analyses.
# male_vs_female_authors_analysis_dunning('gutenberg')
# he_vs_she_associations_analysis_dunning('gutenberg')
# female_characters_author_gender_differences('gutenberg')
# male_characters_author_gender_differences('gutenberg')
# god_author_gender_differences('gutenberg')
# money_author_gender_differences('gutenberg')
# dunning_result_to_dict(male_vs_female_authors_analysis_dunning('gutenberg'))
from dh_testers.testRunner import main_test
main_test()
You can’t perform that action at this time.