def get_fk_grade_level(text):
# The text must contain at least 100 words
if len(text.split()) < 100:
result = "ERROR: This piece of text is too short to get a Flesch Kincaid grade level."
else:
# Instantiate a Readability object
r = Readability(text)
# Get the F-K score metric
fk = r.flesch_kincaid()
# Get the F-K grade level
result = fk.grade_level
return result
def doc_to_readability(doc_str) -> ArrayField:
if len(doc_str) < 10:
return ArrayField(np.zeros(7))
str_to_read = doc_str
try:
while len(str_to_read.split()) < 150:
str_to_read += " " + doc_str
r = Readability(str_to_read)
r_scores = [
r.flesch_kincaid().score,
r.flesch().score,
r.gunning_fog().score,
r.coleman_liau().score,
r.dale_chall().score,
r.ari().score,
r.linsear_write().score
]
return ArrayField(np.array(r_scores))
except ReadabilityException:
return ArrayField(np.zeros(7))
def run(self, book, **kwargs):
doc = book.plaintext
isbn = 'isbn' in book.metadata and book.metadata['isbn'][0]
url = 'https://atlas-fab.lexile.com/free/books/' + str(isbn)
headers = {'accept': 'application/json; version=1.0'}
lexile = requests.get(url, headers=headers)
# Checks if lexile exists for ISBN. If doesn't exist value remains 'None'.
# If lexile does exist but no age range, value will be 'None'.
# If no ISBN, value will be 'None'.
if lexile.status_code == 200:
lexile_work = lexile.json()['data']['work']
self.lexile_min_age = str(lexile_work['min_age'])
self.lexile_max_age = str(lexile_work['max_age'])
try:
r = Readability(doc)
fk = r.flesch_kincaid()
s = r.smog()
self.readability_fk_score = fk.score
self.readability_s_score = s.score
# If less than 100 words
except ReadabilityException:
pass
import requests
from bs4 import BeautifulSoup
from readability import Readability
response = requests.get('http://127.0.0.1/demo')
rtext = "The result is a number that corresponds with a U.S. grade level. The sentence, The Australian platypus is seemingly a hybrid of a mammal and reptilian creature is an 11.3 as it has 24 syllables and 13 words. The different weighting factors for words per sentence and syllables per word in each scoring system mean that the two schemes are not directly comparable and cannot be converted. The grade level formula emphasises sentence length over word length. By creating one-word strings with hundreds of random characters, grade levels may be attained that are hundreds of times larger than high school completion in the United States. Due to the formula's construction, the score does not have an upper bound."
robj = Readability(rtext)
html = response.text
obj = BeautifulSoup(html,'lxml')
def getTagCount(c):
ct = obj.find_all(c)
print("Count of '"+c+"' Tag is :" + str(len(ct)))
print ct
if response.status_code == 200:
print('Connection Success!\n')
elif response.status_code == 404:
print('Not Found.\n')
getTagCount('a')
#print len(rtext.split())
fk = robj.flesch_kincaid()
s = robj.smog()
print "\nFlesch Kincaid Readabiity Score : "+str(fk.score)
print "Flesch Kincaid Readabiity Grade Level : "+str(fk.grade_level)
#print "SMOG Readabiity Score : "+str(s.score)
#print "SMOG Readabiity Grade Level : "+str(s.grade_level)
def flesch_kincaid_score(essay):
r = Readability(essay)
f = r.flesch_kincaid()
return f.score
class ReadabilityTest(unittest.TestCase):
def setUp(self):
text = """
“On a June day sometime in the early 1990s, encouraged by his friend and fellow economist Jörgen Weibull, Abhijit went swimming in the Baltic. He leaped in and instantly jumped out—he claims that his teeth continued to chatter for the next three days. In 2018, also in June, we went to the Baltic in Stockholm, several hundred miles farther north than the previous encounter. This time it was literally child’s play; our children frolicked in the water.
Wherever we went in Sweden, the unusually warm weather was a topic of conversation. It was probably a portent of something everyone felt, but for the moment it was hard not to be quite delighted with the new opportunities for outdoor life it offered.”.
"""
self.readability = Readability(text)
def test_ari(self):
r = self.readability.ari()
print(r)
self.assertEqual(9.551245421245422, r.score)
self.assertEqual(['10'], r.grade_levels)
self.assertEqual([15, 16], r.ages)
def test_coleman_liau(self):
r = self.readability.coleman_liau()
print(r)
self.assertEqual(10.673162393162393, r.score)
self.assertEqual('11', r.grade_level)
def test_dale_chall(self):
r = self.readability.dale_chall()
print(r)
self.assertEqual(9.32399010989011, r.score)
self.assertEqual(['college'], r.grade_levels)
def test_flesch(self):
r = self.readability.flesch()
print(r)
self.assertEqual(51.039230769230784, r.score)
self.assertEqual(['10', '11', '12'], r.grade_levels)
self.assertEqual('fairly_difficult', r.ease)
def test_flesch_kincaid(self):
r = self.readability.flesch_kincaid()
print(r)
self.assertEqual(10.125531135531137, r.score)
self.assertEqual('10', r.grade_level)
def test_gunning_fog(self):
r = self.readability.gunning_fog()
print(r)
self.assertEqual(12.4976800976801, r.score)
self.assertEqual('12', r.grade_level)
def test_linsear_write(self):
r = self.readability.linsear_write()
print(r)
self.assertEqual(11.214285714285714, r.score)
self.assertEqual('11', r.grade_level)
def test_smog(self):
text = """
“On a June day sometime in the early 1990s, encouraged by his friend and fellow economist Jörgen Weibull, Abhijit went swimming in the Baltic. He leaped in and instantly jumped out—he claims that his teeth continued to chatter for the next three days. In 2018, also in June, we went to the Baltic in Stockholm, several hundred miles farther north than the previous encounter. This time it was literally child’s play; our children frolicked in the water.
Wherever we went in Sweden, the unusually warm weather was a topic of conversation. It was probably a portent of something everyone felt, but for the moment it was hard not to be quite delighted with the new opportunities for outdoor life it offered.”.
"""
text = ' '.join(text for i in range(0, 5))
readability = Readability(text)
#Test SMOG with 30 sentences
r1 = readability.smog()
#Test SMOG with all sentences
r2 = readability.smog(all_sentences=True)
print("all_sentences=False: %s ; all_sentences=True: %s" % (r1, r2))
self.assertEqual(12.516099999999998, r1.score)
self.assertEqual('13', r1.grade_level)
self.assertEqual(12.785403640627713, r2.score)
self.assertEqual('13', r2.grade_level)
def test_spache(self):
r = self.readability.spache()
print(r)
self.assertEqual(7.164945054945054, r.score)
self.assertEqual('7', r.grade_level)
def test_print_stats(self):
stats = self.readability.statistics()
self.assertEqual(562, stats['num_letters'])
self.assertEqual(117, stats['num_words'])
self.assertEqual(7, stats['num_sentences'])
self.assertEqual(20, stats['num_polysyllabic_words'])
class ReadabilityAnalyser:
def __init__(self, text):
self.readability = Readability(text)
self.FLESCH_KINCAID = ['score', 'grade_level']
self.FLESCH_EASE = ['score', 'ease', 'grade_level']
self.DALE_CHALL = ['score', 'grade_level']
self.ARI = ['score', 'grade_level', 'ages']
self.CLI = ['score', 'grade_level']
self.GUNNING_FOG = ['score', 'grade_level']
self.SMOG = ['score', 'grade_level']
self.SPACHE = ['score', 'grade_level']
self.LINSEAR_WRITE = ['score', 'grade_level']
self.values_index = self.initialize_value_index_array()
def initialize_value_index_array(self):
values_index = dict()
values_index["flesch_kincaid"] = self.FLESCH_KINCAID
values_index["flesch_ease"] = self.FLESCH_EASE
values_index["dale_chall"] = self.DALE_CHALL
values_index["ari"] = self.ARI
values_index["cli"] = self.CLI
values_index["gunning_fog"] = self.GUNNING_FOG
values_index["smog_all"] = self.SMOG
values_index["smog"] = self.SMOG
values_index["spache"] = self.SPACHE
values_index["linsear_write"] = self.LINSEAR_WRITE
return values_index
def flesch_kincaid(self, content, error_ignore=True):
try:
record = dict()
fk = self.readability.flesch_kincaid()
record['score'] = fk.score
record['grade_level'] = fk.grade_level
content["flesch_kincaid"] = record
except ReadabilityException as e:
if not error_ignore:
content["flesch_kincaid"] = str(e)
print(e)
def flesch_ease(self, content, error_ignore=True):
try:
record = dict()
flesch_ease = self.readability.flesch()
record['score'] = flesch_ease.score
record['ease'] = flesch_ease.ease
record['grade_levels'] = flesch_ease.grade_levels
content['flesch_ease'] = record
except ReadabilityException as e:
if not error_ignore:
content['flesch_ease'] = str(e)
print(e)
def dale_chall(self, content, error_ignore=True):
try:
record = dict()
dale_chall = self.readability.dale_chall()
record['score'] = dale_chall.score
record['grade_level'] = dale_chall.grade_levels
content['dale_chall'] = record
except ReadabilityException as e:
if not error_ignore:
content['dale_chall'] = str(e)
print(e)
def automated_readability_index(self, content, error_ignore=True):
try:
record = dict()
ari = self.readability.ari()
record['score'] = ari.score
record['grade_level'] = ari.grade_levels
record['ages'] = ari.ages
content['ari'] = record
except ReadabilityException as e:
if not error_ignore:
content['ari'] = str(e)
print(e)
def coleman_liau_index(self, content, error_ignore=True):
try:
record = dict()
coleman_liau = self.readability.coleman_liau()
record['score'] = coleman_liau.score
record['grade_level'] = coleman_liau.grade_level
content['cli'] = record
print(record)
except ReadabilityException as e:
print(e)
if not error_ignore:
content['cli'] = str(e)
print(e)
def gunning_fog_index(self, content, error_ignore=True):
try:
record = dict()
gunning_fog = self.readability.gunning_fog()
record['score'] = gunning_fog.score
record['grade_level'] = gunning_fog.grade_level
content['gunning_fog'] = record
except ReadabilityException as e:
if not error_ignore:
content['gunning_fog'] = str(e)
print(e)
def smog(self, content, all_sentences=False, error_ignore=True):
record = dict()
try:
if all_sentences:
smog = self.readability.smog(all_sentences=all_sentences)
record['score'] = smog.score
record['grade_level'] = smog.grade_level
content['smog_all'] = record
else:
smog = self.readability.smog()
record['score'] = smog.score
record['grade_level'] = smog.grade_level
content['smog'] = record
except ReadabilityException as e:
print(e)
print(error_ignore)
if not error_ignore:
if all_sentences:
content['smog_all'] = str(e)
else:
content['smog'] = str(e)
print(e)
def spache_readability_formula(self, content, error_ignore=True):
try:
record = dict()
spache = self.readability.spache()
record['score'] = spache.score
record['grade_level'] = spache.grade_level
content['spache'] = record
except ReadabilityException as e:
if not error_ignore:
content['spache'] = str(e)
print(e)
def linsear_write(self, content, error_ignore=True):
try:
record = dict()
linsear_write = self.readability.linsear_write()
record['score'] = linsear_write.score
record['grade_level'] = linsear_write.grade_level
content['linsear_write'] = record
except ReadabilityException as e:
if not error_ignore:
content['linsear_write'] = str(e)
print(e)
@staticmethod
def check_readability_from_file(input_json_file, output_json_file):
result = []
json_file = load_as_json(input_json_file)
for record in json_file:
analyser = ReadabilityAnalyser(record['text'])
analysed_file_record = dict()
analysed_file_record['file'] = record['file']
analysed_file_record['category'] = record['category']
analyser.flesch_kincaid(analysed_file_record)
analyser.flesch_ease(analysed_file_record)
analyser.dale_chall(analysed_file_record)
analyser.automated_readability_index(analysed_file_record)
analyser.coleman_liau_index(analysed_file_record)
analyser.gunning_fog_index(analysed_file_record)
analyser.smog(analysed_file_record)
analyser.smog(analysed_file_record, True)
analyser.spache_readability_formula(analysed_file_record)
analyser.linsear_write(analysed_file_record)
result.append(analysed_file_record)
save_as_json(result, output_json_file)
def check_readability(self, use_methods=None, errors_included=True):
result_analysis = dict()
if use_methods is None or 'flesch_kincaid' in use_methods:
self.flesch_kincaid(result_analysis,
error_ignore=not errors_included)
if use_methods is None or 'flesch_ease' in use_methods:
self.flesch_ease(result_analysis, error_ignore=not errors_included)
if use_methods is None or 'dale_chall' in use_methods:
self.dale_chall(result_analysis, error_ignore=not errors_included)
if use_methods is None or 'ari' in use_methods:
self.automated_readability_index(result_analysis,
error_ignore=not errors_included)
if use_methods is None or 'cli' in use_methods:
self.coleman_liau_index(result_analysis,
error_ignore=not errors_included)
if use_methods is None or 'gunning_fog' in use_methods:
self.gunning_fog_index(result_analysis,
error_ignore=not errors_included)
if use_methods is None or 'smog' in use_methods:
self.smog(result_analysis, error_ignore=not errors_included)
if use_methods is None or 'smog_all' in use_methods:
self.smog(result_analysis, True, error_ignore=not errors_included)
if use_methods is None or 'spache' in use_methods:
self.spache_readability_formula(result_analysis,
error_ignore=not errors_included)
if use_methods is None or 'linsear_write' in use_methods:
self.linsear_write(result_analysis,
error_ignore=not errors_included)
return result_analysis
@staticmethod
def initialize_basic_dict(categories, values, process_category=True):
record = dict()
for value in values:
record = ReadabilityAnalyser.initialize_dict(record, value)
if process_category:
for category in categories:
record[category] = ReadabilityAnalyser.initialize_basic_dict(
categories, values, False)
return record
@staticmethod
def initialize_dict(record, value):
record['min_' + value] = 999999999
record['max_' + value] = -999999999
record['sum_' + value] = 0
record['avg_' + value] = 0
record['freq_' + value] = 0
record['skipped_' + value] = 0
return record
def initialize_values(self, statistic, categories):
statistic["flesch_kincaid"] = self.initialize_basic_dict(
categories, self.FLESCH_KINCAID)
statistic["flesch_ease"] = self.initialize_basic_dict(
categories, self.FLESCH_EASE)
statistic["dale_chall"] = self.initialize_basic_dict(
categories, self.DALE_CHALL)
statistic["ari"] = self.initialize_basic_dict(categories, self.ARI)
statistic["cli"] = self.initialize_basic_dict(categories, self.CLI)
statistic["gunning_fog"] = self.initialize_basic_dict(
categories, self.GUNNING_FOG)
statistic["smog_all"] = self.initialize_basic_dict(
categories, self.SMOG)
statistic["smog"] = self.initialize_basic_dict(categories, self.SMOG)
statistic["spache"] = self.initialize_basic_dict(
categories, self.SPACHE)
statistic["linsear_write"] = self.initialize_basic_dict(
categories, self.LINSEAR_WRITE)
statistic['indexes'] = [
"flesch_kincaid", "flesch_ease", "dale_chall", "ari", "cli",
"gunning_fog", "smog_all", "smog", "spache", "linsear_write"
]
statistic['categories'] = categories
@staticmethod
def fill_min_max_sum_category(index, value_index, statistics,
readability_index, category):
if index[value_index] < statistics[readability_index][category][
'min_' + value_index]:
statistics[readability_index][category][
'min_' + value_index] = index[value_index]
if index[value_index] > statistics[readability_index][category][
'max_' + value_index]:
statistics[readability_index][category][
'max_' + value_index] = index[value_index]
statistics[readability_index][category]['sum_' + value_index] = \
statistics[readability_index][category]['sum_' + value_index] + index[value_index]
@staticmethod
def fill_min_max_sum_category_value(value, value_index, statistics,
readability_index, category):
if value < statistics[readability_index][category]['min_' +
value_index]:
statistics[readability_index][category]['min_' +
value_index] = value
if value > statistics[readability_index][category]['max_' +
value_index]:
statistics[readability_index][category]['max_' +
value_index] = value
statistics[readability_index][category]['sum_' + value_index] = \
statistics[readability_index][category]['sum_' + value_index] + value
@staticmethod
def fill_min_max_sum(index, value_index, statistics, readability_index):
if index[value_index] < statistics[readability_index]['min_' +
value_index]:
statistics[readability_index]['min_' +
value_index] = index[value_index]
if index[value_index] > statistics[readability_index]['max_' +
value_index]:
statistics[readability_index]['max_' +
value_index] = index[value_index]
statistics[readability_index]['sum_' + value_index] = \
statistics[readability_index]['sum_' + value_index] + index[value_index]
@staticmethod
def fill_min_max_sum_value(value, value_index, statistics,
readability_index):
if value < statistics[readability_index]['min_' + value_index]:
statistics[readability_index]['min_' + value_index] = value
if value > statistics[readability_index]['max_' + value_index]:
statistics[readability_index]['max_' + value_index] = value
statistics[readability_index]['sum_' + value_index] = \
statistics[readability_index]['sum_' + value_index] + value
@staticmethod
def cast_to_float(value):
try:
return float(value)
except ValueError:
return None
except TypeError:
return None
def record_analysis(self, record, statistics):
for readability_index in statistics['indexes']:
if 'category' in record:
category = record['category']
if readability_index in record:
index = record[readability_index]
for value_index in self.values_index[readability_index]:
if value_index in index:
obtained_value = ReadabilityAnalyser.cast_to_float(
index[value_index])
if obtained_value is not None:
index[value_index] = obtained_value
ReadabilityAnalyser.fill_min_max_sum_category(
index, value_index, statistics,
readability_index, category)
if 'freq_' + value_index not in statistics[
readability_index][category]:
statistics[readability_index][category][
'freq_' + value_index] = 0
statistics[readability_index][category]['freq_' + value_index] = \
statistics[readability_index][category]['freq_' + value_index] + 1
elif isinstance(index[value_index], list):
for rec in index[value_index]:
if value_index not in statistics[
readability_index][category]:
statistics[readability_index][
category][value_index] = dict()
if isinstance(rec, str):
if 'freq_' + rec not in statistics[
readability_index][category][
value_index]:
statistics[readability_index][
category][value_index]['freq_'
+
rec] = 0
statistics[readability_index][category][value_index]['freq_' + rec] = \
statistics[readability_index][category][value_index]['freq_' + rec] + 1
else:
ReadabilityAnalyser.fill_min_max_sum_category_value(
rec, value_index, statistics,
readability_index, category)
if 'freq_' + value_index not in \
statistics[readability_index][category][value_index]:
statistics[readability_index][
category]['freq_' +
value_index] = 0
statistics[readability_index][category]['freq_' + value_index] = \
statistics[readability_index][category]['freq_' + value_index] + 1
elif isinstance(index[value_index], str):
rec = index[value_index]
if value_index not in statistics[
readability_index][category]:
statistics[readability_index][category][
value_index] = dict()
if 'freq_' + rec not in statistics[
readability_index][category][
value_index]:
statistics[readability_index][category][
value_index]['freq_' + rec] = 0
statistics[readability_index][category][value_index]['freq_' + rec] = \
statistics[readability_index][category][value_index]['freq_' + rec] + 1
else:
print("Uncategorized: " +
str(index[value_index]))
statistics[readability_index][category]['freq_' + value_index] = \
statistics[readability_index][category]['freq_' + value_index] + 1
else:
statistics[readability_index][category]['skipped_' + value_index] = \
statistics[readability_index][category]['skipped_' + value_index] + 1
else:
for value_index in self.values_index[readability_index]:
statistics[readability_index][category]['skipped_' + value_index] = \
statistics[readability_index][category]['skipped_' + value_index] + 1
else:
print("THIS: " + record)
if readability_index in record:
index = record[readability_index]
for value_index in self.values_index[readability_index]:
if value_index in index:
obtained_value = ReadabilityAnalyser.cast_to_float(
index[value_index])
if obtained_value is not None:
index[value_index] = float(index[value_index])
ReadabilityAnalyser.fill_min_max_sum(
index, value_index, statistics,
readability_index)
if 'freq_' + value_index not in statistics[
readability_index]:
statistics[readability_index]['freq_' +
value_index] = 0
statistics[readability_index]['freq_' + value_index] = \
statistics[readability_index]['freq_' + value_index] + 1
elif isinstance(index[value_index], list):
for rec in index[value_index]:
if value_index not in statistics[
readability_index]:
statistics[readability_index][
value_index] = dict()
if isinstance(rec, str):
# print(value_index + " " + str(index[value_index]))
if 'freq_' + rec not in statistics[
readability_index][value_index]:
statistics[readability_index][
value_index]['freq_' + rec] = 0
statistics[readability_index][value_index]['freq_' + rec] = \
statistics[readability_index][value_index]['freq_' + rec] + 1
else:
ReadabilityAnalyser.fill_min_max_sum_value(
rec, value_index, statistics,
readability_index)
if 'freq_' + value_index not in statistics[
readability_index][value_index]:
statistics[readability_index][
'freq_' + value_index] = 0
statistics[readability_index]['freq_' + value_index] = \
statistics[readability_index]['freq_' + value_index] + 1
elif isinstance(index[value_index], str):
rec = index[value_index]
if value_index not in statistics[
readability_index]:
statistics[readability_index][
value_index] = dict()
if 'freq_' + rec not in statistics[
readability_index][value_index]:
statistics[readability_index][value_index][
'freq_' + rec] = 0
statistics[readability_index][value_index]['freq_' + rec] = \
statistics[readability_index][value_index]['freq_' + rec] + 1
else:
print("Uncategorized: " + str(index[value_index]))
statistics[readability_index]['freq_' + value_index] = \
statistics[readability_index]['freq_' + value_index] + 1
else:
statistics[readability_index]['skipped_' + value_index] = \
statistics[readability_index]['skipped_' + value_index] + 1
else:
for value_index in self.values_index[readability_index]:
statistics[readability_index]['skipped_' + value_index] = \
statistics[readability_index]['skipped_' + value_index] + 1
def count_average(self, statistics):
for readability_index in statistics['indexes']:
for value_index in self.values_index[readability_index]:
if statistics[readability_index]['sum_' + value_index] != 0:
statistics[readability_index]['avg_' + value_index] = \
statistics[readability_index]['sum_' + value_index] / statistics[readability_index][
'freq_' + value_index]
else:
statistics[readability_index]['sum_' + value_index] = 0
for category in statistics['categories']:
if statistics[readability_index][category][
'sum_' + value_index] != 0:
statistics[readability_index][category]['avg_' + value_index] = \
statistics[readability_index][category]['sum_' + value_index] / \
statistics[readability_index][category]['freq_' + value_index]
else:
statistics[readability_index][category][
'sum_' + value_index] = 0
def analyse_readability_file(self, readability_file, categories):
statistic = dict()
self.initialize_values(statistic, categories)
file = load_as_json(readability_file)
for record in file:
self.record_analysis(record, statistic)
self.count_average(statistic)
return statistic
def analyse_readability_file_save_results(self, readability_file,
output_statistics_file,
categories):
statistics = self.analyse_readability_file(readability_file,
categories)
save_as_json(statistics, output_statistics_file)
def suggest():
#get language
lang = request.args.get('lang')
if lang == 'en':
#get url
url = request.args.get('url')
#get the html from the URL
import requests
r = requests.get(url)
html = r.text
#get the html content as text - get content from the "main" tag
from bs4 import BeautifulSoup
original_soup = BeautifulSoup(html, features="lxml").find('main')
original_text = original_soup.get_text()
#get initial readability total_score
from readability import Readability
r_o = Readability(original_text)
original_fk = r_o.flesch_kincaid()
#add periods after bullet points and headings so that the Flesch Kicaid score considers them as sentences
html1 = html.replace("</li>", ".</li>")
html2 = html1.replace("</h1>", ".</h1>")
html3 = html2.replace("</h2>", ".</h2>")
html4 = html3.replace("</h3>", ".</h3>")
html5 = html4.replace("</h4>", ".</h4>")
html6 = html5.replace("</h5>", ".</h5>")
html7 = html6.replace("</h6>", ".</h6>")
#get adjusted readability total_score
revised_soup = BeautifulSoup(html7, features="lxml").find('main')
revised_text = revised_soup.get_text()
from readability import Readability
r_f = Readability(revised_text)
final_fk = r_f.flesch_kincaid()
#tokenize the text for processing
import nltk
from nltk.tokenize import RegexpTokenizer
tokenizer = RegexpTokenizer('\w+')
tokens = tokenizer.tokenize(revised_text)
words = []
for word in tokens:
words.append(word.lower())
#remove stop words from the tokens to get only the meaningful words
nltk.download('stopwords')
sw = nltk.corpus.stopwords.words('english')
words_ns = []
for word in words:
if word not in sw:
words_ns.append(word)
#get the 15 most used words in the text
from nltk import FreqDist
fdist1 = FreqDist(words_ns)
most_common = fdist1.most_common(20)
#get all headings and calculate how many words on average between headings
headings = original_soup.findAll(['h1', 'h2', 'h3', 'h4', 'h5', 'h6'])
hratio = len(words) / (len(headings))
#get all paragraphs and all bulleted list, and calculate how many words per paragraph on average
paragraphs = original_soup.findAll(['p', 'ul'])
pratio = (len(words) / len(paragraphs))
#calculate points for readability
if final_fk.score <= 6:
fkpoints = 60
elif final_fk.score >= 18:
fkpoints = 0
else:
fkpoints = (60 - ((final_fk.score - 6) * 5))
#calculate points for number of words between headings
if hratio <= 40:
hpoints = 20
elif hratio >= 200:
hpoints = 0
else:
hpoints = (20 - ((hratio - 40) * 0.125))
#calculate points for number of words per paragraph
if pratio <= 30:
ppoints = 20
elif pratio >= 80:
ppoints = 0
else:
ppoints = (20 - ((pratio - 30) * 0.4))
#add all points
total_score = fkpoints + hpoints + ppoints
total_score = format(total_score, '.2f')
fkpoints = format(fkpoints, '.2f')
final_fk_score = format(final_fk.score, '.2f')
hpoints = format(hpoints, '.2f')
hratio = format(hratio, '.2f')
ppoints = format(ppoints, '.2f')
pratio = format(pratio, '.2f')
total_words = len(words)
return render_template("read_score_en.html",
total_score=total_score,
fkpoints=fkpoints,
final_fk_score=final_fk_score,
hpoints=hpoints,
hratio=hratio,
ppoints=ppoints,
pratio=pratio,
total_words=total_words,
most_common=most_common)
if lang == 'fr':
#get url
url = request.args.get('url')
#get the html from the URL
import requests
r = requests.get(url)
html = r.text
#get the html content as text - get content from the "main" tag
from bs4 import BeautifulSoup
original_soup = BeautifulSoup(html, features="lxml").find('main')
original_text = original_soup.get_text()
#get initial readability total_score
from readability import Readability
r_o = Readability(original_text)
original_fk = r_o.flesch_kincaid()
#add periods after bullet points and headings so that the Flesch Kicaid score considers them as sentences
html1 = html.replace("</li>", ".</li>")
html2 = html1.replace("</h1>", ".</h1>")
html3 = html2.replace("</h2>", ".</h2>")
html4 = html3.replace("</h3>", ".</h3>")
html5 = html4.replace("</h4>", ".</h4>")
html6 = html5.replace("</h5>", ".</h5>")
html7 = html6.replace("</h6>", ".</h6>")
#get adjusted readability total_score
revised_soup = BeautifulSoup(html7, features="lxml").find('main')
revised_text = revised_soup.get_text()
from readability import Readability
r_f = Readability(revised_text)
final_fk = r_f.flesch_kincaid()
#tokenize the text for processing
import nltk
from nltk.tokenize import RegexpTokenizer
tokenizer = RegexpTokenizer('\w+')
tokens = tokenizer.tokenize(revised_text)
words = []
for word in tokens:
words.append(word.lower())
#remove stop words from the tokens to get only the meaningful words
nltk.download('stopwords')
sw = nltk.corpus.stopwords.words('french')
words_ns = []
for word in words:
if word not in sw:
words_ns.append(word)
#get the 15 most used words in the text
from nltk import FreqDist
fdist1 = FreqDist(words_ns)
most_common = fdist1.most_common(20)
#get all headings and calculate how many words on average between headings
headings = original_soup.findAll(['h1', 'h2', 'h3', 'h4', 'h5', 'h6'])
hratio = len(words) / (len(headings))
#get all paragraphs and all bulleted list, and calculate how many words per paragraph on average
paragraphs = original_soup.findAll(['p', 'ul'])
pratio = (len(words) / len(paragraphs))
#calculate points for readability
if final_fk.score <= 6:
fkpoints = 60
elif final_fk.score >= 18:
fkpoints = 0
else:
fkpoints = (60 - ((final_fk.score - 6) * 5))
#calculate points for number of words between headings
if hratio <= 40:
hpoints = 20
elif hratio >= 200:
hpoints = 0
else:
hpoints = (20 - ((hratio - 40) * 0.125))
#calculate points for number of words per paragraph
if pratio <= 30:
ppoints = 20
elif pratio >= 80:
ppoints = 0
else:
ppoints = (20 - ((pratio - 30) * 0.4))
#add all points
total_score = fkpoints + hpoints + ppoints
total_score = format(total_score, '.2f')
fkpoints = format(fkpoints, '.2f')
final_fk_score = format(final_fk.score, '.2f')
hpoints = format(hpoints, '.2f')
hratio = format(hratio, '.2f')
ppoints = format(ppoints, '.2f')
pratio = format(pratio, '.2f')
total_words = len(words)
return render_template("read_score_fr.html",
total_score=total_score,
fkpoints=fkpoints,
final_fk_score=final_fk_score,
hpoints=hpoints,
hratio=hratio,
ppoints=ppoints,
pratio=pratio,
total_words=total_words,
most_common=most_common)
def suggest():
#get language
lang = request.args.get('lang', 'en')
import nltk
nltk.download('punkt')
if lang == 'en':
word_column_names = ['Count', 'Word']
if lang == 'fr':
word_column_names = ['Nombre', 'Mot']
#get url
url = request.args.get('url', 'https://www.canada.ca/en.html')
#get the html from the URL
import requests
r = requests.get(url)
html = r.text
#get the html content as text - get content from the "main" tag
from bs4 import BeautifulSoup
original_soup = BeautifulSoup(html, features="lxml").find('main')
original_text = original_soup.get_text()
original_text = original_text.replace('..', '.')
original_text = original_text.replace('.', '. ')
original_text = original_text[:original_text.find("defPreFooter")]
original_text = original_text.replace('\n', '')
original_text = original_text.replace('\t', '')
original_text = original_text.replace('\r', '')
#get initial readability total_score
from readability import Readability
r_o = Readability(original_text)
original_fk = r_o.flesch_kincaid()
original_score = original_fk.score
original_score = format(original_score, '.2f')
#add periods after bullet points and headings so that the Flesch Kicaid score considers them as sentences
html1 = html.replace("</li>", ".</li>")
html2 = html1.replace("</h1>", ".</h1>")
html3 = html2.replace("</h2>", ".</h2>")
html4 = html3.replace("</h3>", ".</h3>")
html5 = html4.replace("</h4>", ".</h4>")
html6 = html5.replace("</h5>", ".</h5>")
html7 = html6.replace("</h6>", ".</h6>")
#get adjusted readability total_score
revised_soup = BeautifulSoup(html7, features="lxml").find('main')
for t in revised_soup.select('table'):
t.extract()
revised_text = revised_soup.get_text()
revised_text = revised_text.replace('..', '.')
revised_text = revised_text .replace('.', '. ')
revised_text = revised_text[:revised_text.find("defPreFooter")]
revised_text = revised_text.replace('\n', '')
revised_text = revised_text.replace('\t', '')
revised_text = revised_text.replace('\r', '')
from readability import Readability
r_f = Readability(revised_text)
final_fk = r_f.flesch_kincaid()
#tokenize the text for processing
from nltk.tokenize import RegexpTokenizer
tokenizer = RegexpTokenizer('\w+')
tokens = tokenizer.tokenize(revised_text)
words = []
for word in tokens:
words.append(word.lower())
#remove stop words from the tokens to get only the meaningful words
nltk.download('stopwords')
sw_en = nltk.corpus.stopwords.words('english')
words_ns_en = []
for word in words:
if word not in sw_en:
words_ns_en.append(word)
#get the 15 most used words in the text
from nltk import FreqDist
fdist1_en = FreqDist(words_ns_en)
most_common_en = fdist1_en.most_common(20)
mc_en = pd.DataFrame(most_common_en, columns =['Word', 'Count'])
mc_en = mc_en[['Count', 'Word']]
sw_fr = nltk.corpus.stopwords.words('french')
words_ns_fr = []
for word in words:
if word not in sw_fr:
words_ns_fr.append(word)
#get the 15 most used words in the text
from nltk import FreqDist
fdist1_fr = FreqDist(words_ns_fr)
most_common_fr = fdist1_fr.most_common(20)
mc_fr = pd.DataFrame(most_common_fr, columns =['Mot', 'Nombre'])
#get all headings and calculate how many words on average between headings
headings = original_soup.findAll(['h1', 'h2', 'h3', 'h4', 'h5', 'h6'])
len_headings = len(headings)
hratio = len(words)/(len(headings))
#get all paragraphs and all bulleted list, and calculate how many words per paragraph on average
paragraphs = original_soup.findAll(['p', 'ul'])
len_par = len(paragraphs)
pratio = (len(words)/len(paragraphs))
#calculate points for readability
if final_fk.score <= 6:
fkpoints = 60
elif final_fk.score >= 18:
fkpoints = 0
else :
fkpoints = (60-((final_fk.score-6)*5))
#calculate points for number of words between headings
if hratio <= 40:
hpoints = 20
elif hratio >= 200:
hpoints = 0
else :
hpoints = (20-((hratio-40)*0.125 ))
#calculate points for number of words per paragraph
if pratio <= 30:
ppoints = 20
elif pratio >= 80:
ppoints = 0
else :
ppoints = (20-((pratio-30)*0.4))
#add all points
total_score = fkpoints+hpoints+ppoints
total_score = format(total_score, '.2f')
fkpoints = format(fkpoints, '.2f')
final_fk_score = format(final_fk.score, '.2f')
hpoints = format(hpoints, '.2f')
hratio = format(hratio, '.2f')
ppoints = format(ppoints, '.2f')
pratio = format(pratio, '.2f')
total_words = len(words)
total_score = float(total_score)
if total_score >= 90:
if lang=='en':
score = 'Outstanding!'
if lang=='fr':
score = 'Excellent!'
elif total_score >= 80 and total_score < 90:
if lang=='en':
score = 'Very good!'
if lang=='fr':
score = 'Très bien!'
elif total_score >= 70 and total_score < 80:
if lang=='en':
score = 'Not too bad'
if lang=='fr':
score = 'Pas mal'
elif total_score >= 60 and total_score < 70:
if lang=='en':
score = 'Needs work'
if lang=='fr':
score = 'À travailler'
elif total_score >= 50 and total_score < 60:
if lang=='en':
score = 'Needs a lot of work'
if lang=='fr':
score = 'Besoin de beaucoup de travail'
elif total_score < 50:
if lang=='en':
score = "Please don't do this to people..."
if lang=='fr':
score = "S'il vous plaît, il faut faire quelque chose..."
if lang == "en":
return render_template("read_score_en.html", total_score = total_score, fkpoints = fkpoints, final_fk_score = final_fk_score, hpoints = hpoints, hratio = hratio, ppoints = ppoints, pratio = pratio, total_words = total_words, url = url, lang = lang, word_column_names = word_column_names, row_data_word_en = list(mc_en.values.tolist()), row_data_word_fr = list(mc_fr.values.tolist()), zip = zip, score = score, len_headings = len_headings, len_par = len_par, original_score = original_score)
if lang == "fr":
return render_template("read_score_fr.html", total_score = total_score, fkpoints = fkpoints, final_fk_score = final_fk_score, hpoints = hpoints, hratio = hratio, ppoints = ppoints, pratio = pratio, total_words = total_words, url = url, lang = lang, word_column_names = word_column_names, row_data_word_en = list(mc_en.values.tolist()), row_data_word_fr = list(mc_fr.values.tolist()), zip = zip, score = score, len_headings = len_headings, len_par = len_par, original_score = original_score)
from readability import Readability
text = open('C:\\...\\ch15_MLK-IHaveADream.txt')
text_up = text.read()
r = Readability(text_up)
flesch_kincaidR = r.flesch_kincaid()
print('The text has a grade ' + flesch_kincaidR.grade_level +
' readability level.')
from readability import Readability
path = "IAEA_output"
str = ""
with open(path, encoding="utf-8") as f:
for line in f.readlines():
str = line
r = Readability(str)
fk = r.flesch_kincaid()
print(fk.score)
print(fk.grade_level)
s = r.smog()
print(s.score)
print(s.grade_level)
dc = r.dale_chall()
print(dc.score)
print(dc.grade_levels)
cl = r.coleman_liau()
print(cl.score)
print(cl.grade_level)
gf = r.gunning_fog()
print(gf.score)
print(gf.grade_level)
# lw = r.linsear_write()
# print(lw.score)
# print(lw.grade_level)
import pandas as pd
from readability import Readability
import sys
import csv
import json
csv.field_size_limit(sys.maxsize)
description_features = {}
with open(
"detail_description_data/detail-desc-text-{}.tsv".format(sys.argv[1]),
'r') as tsvin:
tsvin = csv.reader(tsvin, delimiter='\t')
for row in tsvin:
patent_id = row[0]
detail_description_text = row[1]
try:
description_word_count = len(detail_description_text.split())
fig_counts = detail_description_text.lower().count(
"fig.") + detail_description_text.lower().count("figs.")
r = Readability(detail_description_text.replace('aed-512', ''))
fk_score = r.flesch_kincaid().score
except:
continue
description_features[patent_id] = (description_word_count, fk_score,
fig_counts)
with open(
"feature_data/description_features/description_features_{}.json".
format(sys.argv[1]), 'w') as f:
json.dump(description_features, f)
l_flesch = []
l_gunning_fog = []
l_coleman_liau = []
l_dale_chall = []
l_ari = []
l_linsear_write = []
l_spache = []
l_flesch_ease = []
for i in os.listdir(PATH):
if not i.startswith('.'):
if i not in l_not_use:
with open(PATH + i, 'r') as f:
text = f.read()
r = Readability(clean(text))
s1 = r.flesch_kincaid()
s2 = r.flesch()
s3 = r.gunning_fog()
s4 = r.coleman_liau()
s5 = r.dale_chall()
s6 = r.ari()
s7 = r.linsear_write()
# r.smog()
s8 = r.spache()
l_flesch_kincaid.append(s1.score)
l_flesch.append(s2.score)
l_flesch_ease.append(s2.ease)
l_gunning_fog.append(s3.score)
l_coleman_liau.append(s4.score)
l_dale_chall.append(s5.score)
l_ari.append(s6.score)
def __flesch_kincaid(r: Readability) -> float:
try:
lvl = r.flesch_kincaid().grade_level
return float(lvl)
except ReadabilityException:
return None
class ReadabilityTest(unittest.TestCase):
def setUp(self):
text = """
In linguistics, the Gunning fog index is a readability test for English writing. The index estimates the years of formal education a person needs to understand the text on the first reading. For instance, a fog index of 12 requires the reading level of a United States high school senior (around 18 years old). The test was developed in 1952 by Robert Gunning, an American businessman who had been involved in newspaper and textbook publishing.
The fog index is commonly used to confirm that text can be read easily by the intended audience. Texts for a wide audience generally need a fog index less than 12. Texts requiring near-universal understanding generally need an index less than 8.
"""
self.readability = Readability(text)
def test_ari(self):
r = self.readability.ari()
print(r)
self.assertEqual(9.551245421245422, r.score)
self.assertEqual(['10'], r.grade_levels)
self.assertEqual([15, 16], r.ages)
def test_coleman_liau(self):
r = self.readability.coleman_liau()
print(r)
self.assertEqual(10.673162393162393, r.score)
self.assertEqual('11', r.grade_level)
def test_dale_chall(self):
r = self.readability.dale_chall()
print(r)
self.assertEqual(9.32399010989011, r.score)
self.assertEqual(['college'], r.grade_levels)
def test_flesch(self):
r = self.readability.flesch()
print(r)
self.assertEqual(51.039230769230784, r.score)
self.assertEqual(['10', '11', '12'], r.grade_levels)
self.assertEqual('fairly_difficult', r.ease)
def test_flesch_kincaid(self):
r = self.readability.flesch_kincaid()
print(r)
self.assertEqual(10.125531135531137, r.score)
self.assertEqual('10', r.grade_level)
def test_gunning_fog(self):
r = self.readability.gunning_fog()
print(r)
self.assertEqual(12.4976800976801, r.score)
self.assertEqual('12', r.grade_level)
def test_linsear_write(self):
r = self.readability.linsear_write()
print(r)
self.assertEqual(11.214285714285714, r.score)
self.assertEqual('11', r.grade_level)
def test_smog(self):
text = """
In linguistics, the Gunning fog index is a readability test for English writing. The index estimates the years of formal education a person needs to understand the text on the first reading. For instance, a fog index of 12 requires the reading level of a United States high school senior (around 18 years old). The test was developed in 1952 by Robert Gunning, an American businessman who had been involved in newspaper and textbook publishing.
The fog index is commonly used to confirm that text can be read easily by the intended audience. Texts for a wide audience generally need a fog index less than 12. Texts requiring near-universal understanding generally need an index less than 8.
"""
text = ' '.join(text for i in range(0, 5))
readability = Readability(text)
r = readability.smog()
print(r)
self.assertEqual(12.516099999999998, r.score)
self.assertEqual('13', r.grade_level)
def test_spache(self):
r = self.readability.spache()
print(r)
self.assertEqual(7.164945054945054, r.score)
self.assertEqual('7', r.grade_level)
def test_print_stats(self):
stats = self.readability.statistics()
self.assertEqual(562, stats['num_letters'])
self.assertEqual(117, stats['num_words'])
self.assertEqual(7, stats['num_sentences'])
self.assertEqual(20, stats['num_polysyllabic_words'])
#------------------
# Readability
#------------------
st.header('Readability')
# Context
passage = st.text_area("Candidate Bible Passage (English)", value='',
max_chars=None, key='readability_passage')
# Calculate readability
r = Readability(passage)
# Display readability
data = [
['Flesch-Kincaid Score', r.flesch_kincaid().score],
['Flesch Reading Ease', r.flesch().ease],
['Dale Chall Readability Score', r.dale_chall().score],
['Automated Readability Index Score', r.ari().score],
['Coleman Liau Index', r.coleman_liau().score],
['Gunning Fog', r.gunning_fog().score],
['Linsear Write', r.linsear_write().score],
['Spache Readability Formula', r.spache().score]
]
df = pd.DataFrame(data, columns=['Readability Metric', 'Value'])
if st.button('Assess Readability', key=None):
st.write(df)
def generate_caption_stats(dataframe: pd.DataFrame,
pos_tag_stats: bool = True,
readability_scores: bool = True,
n_spacy_workers: int = 6,
spacy_model: str = "en_core_web_lg",
backend: MetadataGeneratorBackend = MetadataGeneratorBackend.SPACY):
logger.info(f"Generating caption statistics using {backend.upper()}...")
start = time.time()
# Tokens and sentences
num_tok = []
num_sent = []
# Min and Max length of sentences
min_sent_len = []
max_sent_len = []
# Named Entities
num_ne = []
ne_texts = [] # surface form of the NEs
ne_types = [] # types of the NEs
# POS Tags
# counts
num_noun = [] # nouns (cat, dog, house, tree, ...)
num_propn = [] # proper nouns (Denver, Hamburg, Peter, Tesla, ...)
num_conj = [] # conjunctions (and, or, ...)
num_verb = [] # verbs
num_sym = [] # symbols (!,#,?, ...)
num_num = [] # numbers (IV, 1 billion, 1312, ...)
num_adp = [] # adpositions (on, under, in, at, ...)
num_adj = [] # adjectives (nice, fast, cool, ...)
# ratios
ratio_ne_tokens, num_ne_tok = [], []
ratio_noun_tokens = []
ratio_propn_tokens = []
ratio_all_noun_tokens = []
# readability scores
fk_gl_score = []
fk_re_score = []
dc_score = []
with tqdm(total=len(dataframe)) as pbar:
# TODO extract all of this code into an own module and have separate metadata generators for spaCy, nltk, etc.
if backend == MetadataGeneratorBackend.SPACY:
# init spacy TODO: download the required model(s)
spacy_nlp = spacy.load(spacy_model)
if readability_scores:
spacy_nlp.add_pipe(Readability())
# TODO whats a good batch_size?
for doc in spacy_nlp.pipe(dataframe['caption'].astype(str),
n_process=n_spacy_workers):
# num tokens
num_tok.append(len(doc))
# num sentences
num_sent.append(len(list(doc.sents)))
# min/max length of sentences
min_len = 10000
max_len = -1
for s in doc.sents:
min_len = min(min_len, len(s))
max_len = max(max_len, len(s))
min_sent_len.append(min_len)
max_sent_len.append(max_len)
# named entities
num_ne.append(len(doc.ents))
txt, typ = [], []
for ent in doc.ents:
typ.append(ent.label_)
txt.append(ent.text)
ne_texts.append(txt)
ne_types.append(typ)
# readability scores
if readability_scores:
fk_gl_score.append(doc._.flesch_kincaid_grade_level)
fk_re_score.append(doc._.flesch_kincaid_reading_ease)
dc_score.append(doc._.dale_chall)
# POS Tags
if pos_tag_stats:
noun, propn, conj, verb, sym, num, adp, adj, ne_tok = 0, 0, 0, 0, 0, 0, 0, 0, 0
for t in doc:
if t.pos_ == 'CONJ':
conj += 1
elif t.pos_ == 'ADJ':
adj += 1
elif t.pos_ == 'NOUN':
noun += 1
elif t.pos_ == 'NUM':
num += 1
elif t.pos_ == 'PROPN':
propn += 1
elif t.pos_ == 'SYM':
sym += 1
elif t.pos_ == 'VERB':
verb += 1
elif t.pos_ == 'ADP':
adp += 1
# number of tokens associated with a NE (to compute the ratio)
if t.ent_iob_ == 'I' or t.ent_iob_ == 'B':
ne_tok += 1
num_noun.append(noun)
num_propn.append(propn)
num_conj.append(conj)
num_verb.append(verb)
num_sym.append(sym)
num_num.append(num)
num_adp.append(adp)
num_adj.append(adj)
num_ne_tok.append(ne_tok)
pbar.update(1)
elif backend == MetadataGeneratorBackend.NLTK:
nltk.download('punkt')
nltk.download('words')
nltk.download('averaged_perceptron_tagger')
nltk.download('universal_tagset')
nltk.download('universal_treebanks_v20')
nltk.download('maxent_ne_chunker')
for cap in dataframe['caption'].astype(str):
# num tokens
num_tok.append(len(nltk.word_tokenize(cap)))
# num sentences
sents = nltk.sent_tokenize(cap)
num_sent.append(len(sents))
# min/max length of sentences
min_len = 10000
max_len = -1
s_toks = []
for s in sents:
toks = nltk.word_tokenize(s)
s_toks.append(toks)
min_len = min(min_len, len(toks))
max_len = max(max_len, len(toks))
min_sent_len.append(min_len)
max_sent_len.append(max_len)
# readability scores
# FIXME currently not usable with NLTK... (because NaN values are dropped)
# there is also an error while calling t Readability(cap) ctor...
if False:
try:
r = Readability(cap)
flesch = r.flesch_kincaid()
fk_gl_score.append(flesch.grade_level)
fk_re_score.append(flesch.score)
dc_score.append(r.dale_chall().score)
except (ReadabilityException, Exception):
fk_gl_score.append(np.NaN)
fk_re_score.append(np.NaN)
dc_score.append(np.NaN)
if pos_tag_stats:
sent_pos_tags = nltk.pos_tag_sents(s_toks, 'universal')
noun, propn, conj, verb, sym, num, adp, adj, ne_tok = 0, 0, 0, 0, 0, 0, 0, 0, 0
for spt in sent_pos_tags:
for pt in spt:
if pt[1].upper() == 'CONJ':
conj += 1
elif pt[1].upper() == 'ADJ':
adj += 1
elif pt[1].upper() == 'NOUN':
noun += 1
elif pt[1].upper() == 'NUM':
num += 1
elif pt[1].upper() == 'PROPN':
propn += 1
elif pt[1].upper() == 'SYM':
sym += 1
elif pt[1].upper() == 'VERB':
verb += 1
elif pt[1].upper() == 'ADP':
adp += 1
num_noun.append(noun)
num_propn.append(propn)
num_conj.append(conj)
num_verb.append(verb)
num_sym.append(sym)
num_num.append(num)
num_adp.append(adp)
num_adj.append(adj)
# named entities
# we have to tag again with a different tag set (upenn tree) for WAY better NER performance
num_nes, num_nes_tok = 0, 0
txt, typ = [], []
nes_sent = nltk.ne_chunk_sents(nltk.pos_tag_sents(map(nltk.word_tokenize, nltk.sent_tokenize(cap))))
for nes in nes_sent:
for ne in nes:
if isinstance(ne, nltk.Tree):
num_nes += 1
typ.append(str(ne.label()))
t = ""
for tok in ne:
t += tok[0] + " "
txt.append(t.strip())
num_nes_tok += len(ne)
num_ne.append(num_nes)
ne_texts.append(txt)
ne_types.append(typ)
num_ne_tok.append(num_nes_tok)
pbar.update(1)
elif backend == MetadataGeneratorBackend.POLYGLOT:
# init
# pandarallel.initialize() # FIXME doens't work..
downloader.download("embeddings2.en")
downloader.download("ner2.en")
downloader.download("pos2.en")
def __gen_polyglot_metadata_per_caption(df, pb):
d = {
'num_tok': 0,
'num_sent': 0,
'min_sent_len': 0,
'max_sent_len': 0,
'num_ne': 0,
'ne_types': [],
'ne_texts': [],
'num_nouns': 0,
'num_propn': 0,
'num_conj': 0,
'num_verb': 0,
'num_sym': 0,
'num_num': 0,
'num_adp': 0,
'num_adj': 0,
'ratio_ne_tok': 0.,
'ratio_noun_tok': 0.,
'ratio_propn_tok': 0.,
'ratio_all_noun_tok': 0.,
}
try:
caption = str(df['caption']).encode('utf-8')
# https://github.com/aboSamoor/polyglot/issues/71
# removing "bad unicode" characters to avoid runtime exceptions
# caption = str(caption, encoding='utf-8')
caption = regex.sub(r"\p{C}", "", caption.decode('utf-8'))
pg = Text(caption, hint_language_code='en')
pg.language = 'en'
# num tokens
n_tok = len(pg.words)
# num sentences
n_sent = len(pg.sentences)
# min/max length of sentences
min_s_len = 10000
max_s_len = -1
for s in pg.sentences:
min_s_len = min(min_s_len, len(s.words))
max_s_len = max(max_s_len, len(s.words))
# readability scores
# FIXME only available with spacy currently
# POS tags
n_noun, n_propn, n_conj, n_verb, n_sym, n_num, n_adp, n_adj = 0, 0, 0, 0, 0, 0, 0, 0
for pos in pg.pos_tags:
if pos[1].upper() == 'CONJ':
n_conj += 1
elif pos[1].upper() == 'ADJ':
n_adj += 1
elif pos[1].upper() == 'NOUN':
n_noun += 1
elif pos[1].upper() == 'NUM':
n_num += 1
elif pos[1].upper() == 'PROPN':
n_propn += 1
elif pos[1].upper() == 'SYM':
n_sym += 1
elif pos[1].upper() == 'VERB':
n_verb += 1
elif pos[1].upper() == 'ADP':
n_adp += 1
# named entities
num_nes_tok, ne_txt, ne_typ = 0, [], []
num_nes = len(pg.entities)
for ne in pg.entities:
num_nes_tok += len(ne)
ne_txt.append(" ".join(ne))
ne_typ.append(ne.tag)
# compute the rations
r_ne_tokens = num_nes_tok / n_tok
r_noun_tokens = n_noun / n_tok
r_propn_tokens = n_propn / n_tok
r_all_noun_tokens = (n_noun + n_propn) / n_tok
d = {
'num_tok': n_tok,
'num_sent': n_sent,
'min_sent_len': min_s_len,
'max_sent_len': max_s_len,
'num_ne': num_nes,
'ne_types': ne_typ,
'ne_texts': ne_txt,
'num_nouns': n_noun,
'num_propn': n_propn,
'num_conj': n_conj,
'num_verb': n_verb,
'num_sym': n_sym,
'num_num': n_num,
'num_adp': n_adp,
'num_adj': n_adj,
'ratio_ne_tok': r_ne_tokens,
'ratio_noun_tok': r_noun_tokens,
'ratio_propn_tok': r_propn_tokens,
'ratio_all_noun_tok': r_all_noun_tokens,
}
except Exception as e:
logger.error(f"Critical error occurred with caption of WikiCaps ID{df['wikicaps_id']}!")
logger.error(str(e))
return
finally:
pb.update(1)
return d
# FIXME why the hec is this using ALL AVAILABLE CORES?!
metadata = dataframe.apply(__gen_polyglot_metadata_per_caption, axis=1, result_type='expand', args=(pbar,))
res = pd.concat([dataframe, metadata], axis=1)
res.convert_dtypes()
logger.info(f"Finished adding caption statistics in {time.time() - start} seconds!")
return res
# compute the rations
if pos_tag_stats:
np_num_tok = np.array(num_tok)
np_num_noun = np.array(num_noun)
np_num_propn = np.array(num_propn)
ratio_ne_tokens = (np.array(num_ne_tok) / np_num_tok)
ratio_noun_tokens = (np_num_noun / np_num_tok)
ratio_propn_tokens = (np_num_propn / np_num_tok)
ratio_all_noun_tokens = ((np_num_noun + np_num_propn) / np_num_tok)
res = dataframe.copy()
# add stats as columns to df
res['num_tok'] = num_tok
res['num_sent'] = num_sent
res['min_sent_len'] = min_sent_len
res['max_sent_len'] = max_sent_len
res['num_ne'] = num_ne
res['ne_types'] = ne_types
res['ne_texts'] = ne_texts
if pos_tag_stats:
res['num_nouns'] = num_noun
res['num_propn'] = num_propn
res['num_conj'] = num_conj
res['num_verb'] = num_verb
res['num_sym'] = num_sym
res['num_num'] = num_num
res['num_adp'] = num_adp
res['num_adj'] = num_adj
res['ratio_ne_tok'] = ratio_ne_tokens
res['ratio_noun_tok'] = ratio_noun_tokens
res['ratio_propn_tok'] = ratio_propn_tokens
res['ratio_all_noun_tok'] = ratio_all_noun_tokens
if readability_scores:
res['fk_re_score'] = fk_re_score
res['fk_gl_score'] = fk_gl_score
res['dc_score'] = dc_score
res.convert_dtypes() # make sure that ints are not encoded as floats
logger.info(f"Finished adding caption statistics in {time.time() - start} seconds!")
return res
