def test_sent_tokenization(self):
assert len(
tokenize_sents(
"""Результати цих досліджень опубліковано в таких колективних працях, як «Статистичні параметри
стилів», «Морфемна структура слова», «Структурна граматика української мови Проспект», «Частотний словник сучасної української художньої прози», «Закономірності структурної організації науково-реферативного тексту», «Морфологічний аналіз наукового тексту на ЕОМ», «Синтаксичний аналіз наукового тексту на ЕОМ», «Використання ЕОМ у лінгвістичних дослідженнях» та ін. за участю В.І.Перебийніс,
М.М.Пещак, М.П.Муравицької, Т.О.Грязнухіної, Н.П.Дарчук, Н.Ф.Клименко, Л.І.Комарової, В.І.Критської,
Т.К.Пуздирєвої, Л.В.Орлової, Л.А.Алексієнко, Т.І.Недозим.""")) == 1
def tokenize_texts_to_sentences(texts):
text_sentences = []
for text in texts:
text_to_add = []
text_to_add += tokenize_uk.tokenize_sents(text)
text_sentences.append(text_to_add)
return text_sentences
def reformat_split_sents():
with open('../data/scraped.txt', 'r') as inf, \
open('../data/new_scraped.txt', 'w') as outf:
new_sents = [
s for snip in inf.readlines()
for s in tokenize_uk.tokenize_sents(snip)
]
for sent in new_sents:
outf.write(sent + "\n")
def tokenize_sentences(raw, is_tokenize_uk=False):
"""
uses nltk by default
if 'is_tokenize_uk' is True, then uses tokenize_uk
"""
if is_tokenize_uk:
return tokenize_uk.tokenize_sents(raw)
else:
return nltk.sent_tokenize(raw)
def transform_lines_into_rythm(lines, accent_vocab):
rythm = ''
for line in lines.split('\n'):
sent_tokens = tokenize_sents(line)
for sent in sent_tokens:
for word in filter(lambda w: w in accent_vocab,
tokenize_words(sent)):
accent_options = accent_vocab[word]
# use only first one, f**k other options
word, index = accent_options[0]
rythm_map = transform_into_rythm_map(word, index)
rythm += rythm_map
rythm += '\n'
return rythm
def main():
with open('scopus-3.csv',
'rb') as csvfile: #open the citation and abstract data file
citation_set = {}
data_file = csv.reader(csvfile) #create a csv file object
for data in data_file:
if data[1] != "Cited by" and data[
3] != "[No abstract available]": #checking to make sure it isn't the header row
#or a useless row
title = data[0]
if data[1] == "": #dealing with empty citation data
citations = 0
else:
citations = eval(
data[1]) #converting the citations to numbers
abstract = data[3]
citation_set[title] = (
citations, abstract
) #creating a dictionary of titles, citations, and abstracts
data_set = {}
for key in citation_set:
(citation, abstract) = citation_set[key]
try:
text = tokenize_uk.tokenize_sents(
abstract) #prepping the abstract for analysis
read_scores = readability.getmeasures(
text) #analyzing the abstract
data_dict = read_scores[u'readability grades']
data_set[key] = (citation, data_dict[u'FleschReadingEase']
) #getting a chosen metric
except UnicodeDecodeError: #dealing with errors raised by the tokenizer and readability tool
pass
citation_scores = []
readability_scores = []
for key in data_set:
(cit, read) = data_set[key]
citation_scores.append(cit) #creating a list of citation scores
readability_scores.append(read) #creating a list of readability scores
x = numpy.array(readability_scores) #turning the lists into arrays
y = numpy.array(citation_scores)
stat = stats.linregress(x, y) #basic statistical analysis
print stat
def filter_words_with_pos(tokens, useful_tags):
filtered_words = []
for t in tokens:
tag = morph.parse(t)[0].tag.POS
if tag_mapping.get(tag, str(tag)) in useful_tags:
filtered_words.append(t)
return filtered_words
### Preprocessing
tokenized_data = []
for item in dataset:
sents_list = []
sents = tokenize_uk.tokenize_sents(item[1])
for s in sents:
sents_list.append(tokenize_uk.tokenize_words(s))
tokenized_data.append(sents_list)
tokenized_lengths = [len(t) for t in tokenized_data]
print("tokenized")
lemmatized_data = [[lemmatize_tokens(i) for i in t] for t in tokenized_data]
print("lemmatized")
digits_cleared_data = [[[item for item in i if not item.isdigit()] for i in l]
for l in lemmatized_data]
print("digits")
punct_cleared_data = [[[
item for item in i if item not in f'{string.punctuation}”№«»'
] for i in d] for d in digits_cleared_data]
print("punct")
def test_sent_tokenization(self):
assert len(tokenize_sents("""Результати цих досліджень опубліковано в таких колективних працях, як «Статистичні параметри
стилів», «Морфемна структура слова», «Структурна граматика української мови Проспект», «Частотний словник сучасної української художньої прози», «Закономірності структурної організації науково-реферативного тексту», «Морфологічний аналіз наукового тексту на ЕОМ», «Синтаксичний аналіз наукового тексту на ЕОМ», «Використання ЕОМ у лінгвістичних дослідженнях» та ін. за участю В.І.Перебийніс,
М.М.Пещак, М.П.Муравицької, Т.О.Грязнухіної, Н.П.Дарчук, Н.Ф.Клименко, Л.І.Комарової, В.І.Критської,
Т.К.Пуздирєвої, Л.В.Орлової, Л.А.Алексієнко, Т.І.Недозим.""")) == 1
sents_folder = data_folder + 'sents/'
sents_per_chunk = int(7e5) # 700 000
log_interval = 10000
files = [raw_data_folder + 'ukr_lit.txt', raw_data_folder + 'td.txt']
log('starting')
for src_file in files:
with open(src_file, 'rb') as f:
data = f.read()
log('processing file ' + src_file)
text = data.decode('utf-8')
tokens_text = tokenize_uk.tokenize_sents(text)
log('tokenization finished')
sents_number = int(math.ceil(len(tokens_text) / float(sents_per_chunk)))
for i in range(0, sents_number):
sentences = []
chunk = tokens_text[i * sents_per_chunk:(i + 1) * sents_per_chunk]
for sentence in chunk:
sentences.append(tokenize_uk.tokenize_words(sentence))
if items_processed % log_interval == 0:
log('items processed {}'.format(items_processed))
sents_folder = data_folder + 'sents/'
log_interval = 10000
log('starting')
start_time = time.time()
for chunk_file in os.listdir(chunks_folder):
data = pd.read_csv(chunks_folder + chunk_file)
sentences = []
for row in data.itertuples():
title = row[3].decode('utf-8') if isinstance(row[3], basestring) else ''
text = row[2].decode('utf-8') if isinstance(row[2], basestring) else ''
tokens_title = tokenize_uk.tokenize_sents(title)
tokens_text = tokenize_uk.tokenize_sents(text)
for sentence in tokens_title + tokens_text:
sentences.append(tokenize_uk.tokenize_words(sentence))
total_time = time.time() - start_time
if items_processed % log_interval == 0:
log('items processed {}, running time {}'.format(items_processed, total_time))
items_processed += 1
log('saving chunk ')
result_file = os.path.splitext(chunk_file)[0] + '.msg'
def ner_nlp_extracting(text, model, vesum, word2indx, tag2indx, sess, graph):
X = list(
map(
lambda sentence: tokenize_uk.tokenize_words(sentence),
tokenize_uk.tokenize_sents(' '.join(
tokenize_uk.tokenize_words(text)))))
X_tokenized = np.array([[word for word in sentence] for sentence in X])
X = [[
word2indx.get(vesum.get_main_form_from_vesum(word),
word2indx['UNKNOWN']) for word in sentence
] for sentence in X]
X = pad_sequences(X,
maxlen=70,
padding='post',
truncating='post',
value=word2indx['ENDPAD'])
with graph.as_default():
set_session(sess)
pred = np.argmax(model.predict(X), axis=-1)
res = [(sent,
list(
map(
lambda tag: list(
filter(lambda key: tag2indx[key] == tag, tag2indx))[0],
tags[:len(sent)])))
for sent, tags in zip(X_tokenized, pred)]
tokens = list()
tags = list()
for tokens_tmp, tags_tmp in res:
tokens.extend(tokens_tmp)
tags.extend(tags_tmp)
find_tags = list()
start_index = 0
finish_index = 0
for ind, tag in enumerate(tags):
if (ind == 0 or ((ind > 0) and tags[ind - 1] == 'O')) and tag != 'O':
token = tokens[ind]
start_index = text.index(token, finish_index)
finish_index = text.index(token, finish_index) + len(token)
elif tag != 'O':
token = tokens[ind]
finish_index = text.index(token, finish_index) + len(token)
elif ind > 0 and (tags[ind - 1][0] == 'B'
or tags[ind - 1][0] == 'I') and tag == 'O':
ner = tags[ind - 1][2:]
ner_dict = dict()
ner_dict['entity_type'] = ner
ner_dict['start_index'] = start_index
ner_dict['finish_index'] = finish_index
ner_dict['text_entity'] = text[start_index:finish_index]
find_tags.append(ner_dict)
return find_tags
import pymorphy2
import os
import docx2txt
import sqlite3
import tokenize_uk
from string import Template
import time
start = time.time()
texts = []
with os.scandir('C:/Users/user/Desktop/Навчання/Тексти/') as entries:
for entry in entries:
text = docx2txt.process("C:/Users/user/Desktop/Навчання/Тексти/" +
entry.name)
sentences = tokenize_uk.tokenize_sents(text)
texts.append(sentences)
db = sqlite3.connect('dictionary.db')
cursor = db.cursor()
word_forms = {}
count = 1
index = 1
for sents in texts:
for sentence in sents:
sentence = sentence.replace("\n", " ").replace("\"", "")
t = Template(
'INSERT INTO sentences VALUES ($index_of_sent , "$sentence");')
command = t.substitute(index_of_sent=index, sentence=sentence)
cursor.execute(command)
