def lemm_sent_process3(text,
remove_stopwords=False,
summary=False,
mode="nltk",
withdot=False):
nlp = en_core_web_sm.load()
stops = set(stopwords.words("english"))
lines = nltk.sent_tokenize(text)
# Separate out reviewText and summary sentences
text_lines = []
pos_lines = []
for idx, line in enumerate(lines):
if line == "":
continue # empty line
else:
# line = remove_word2(remove_word(line))
# spacy
if mode == "spacy":
# line = remove_word2(remove_word(line))
doc = nlp(line)
# 通過設置doc.is_parsed = True來欺騙spaCy忽略它，即通過讓它相信分配了依賴關係解析並以這種方式應用了句子邊界
# doc.is_parsed = True
sentences = " ".join([
word.lemma_ if word.pos_.startswith('N')
or word.pos_.startswith('J') or word.pos_.startswith('V')
or word.pos_.startswith('R') else word.orth_
for word in doc if ((
not word.is_space and not word.is_bracket
and not word.is_digit and not word.is_left_punct
and not word.is_right_punct and not word.is_bracket
and not word.is_quote and not word.is_currency and
not word.is_punct and word.ent_type_ not in ent_list
and (word.lemma_ not in list(html_escape_table +
stpwords_list2)) and
(word.orth_ not in list(html_escape_table +
stpwords_list2)) and
# not (word.is_stop and remove_stopwords) and
word.tag_ not in ["SYM", "HYPH"]
and word.lemma_ != "-PRON-"
# word.lemma_ not in total_stopwords
))
])
if sentences == "": continue
if withdot == True: sentences = sentences + "."
if summary: sentences = ' <s> ' + sentences + " </s> "
sentences = sentences + "\n"
text_lines.append(sentences)
text_lines = "".join(text_lines)
return text_lines
def lemm_sent_process4(text,
remove_stopwords=False,
summary=False,
mode="nltk",
withdot=False):
stops = set(stopwords.words("english"))
# print('xxx')
lines = nltk.sent_tokenize(text)
# Separate out reviewText and summary sentences
text_lines = []
pos_lines = []
for idx, line in enumerate(lines):
if line == "":
continue # empty line
else:
# line = remove_word2(remove_word(line))
# spacy
if mode == "spacy":
# line = remove_word2(remove_word(line))
doc = nlp(line)
# 通過設置doc.is_parsed = True來欺騙spaCy忽略它，即通過讓它相信分配了依賴關係解析並以這種方式應用了句子邊界
# doc.is_parsed = True
sentences = " ".join([
word.lemma_ if word.pos_.startswith('N')
or word.pos_.startswith('J') or word.pos_.startswith('V')
or word.pos_.startswith('R') else word.orth_
for word in doc if ((
not word.is_space and not word.is_bracket and
# not word.is_digit and
not word.is_left_punct and not word.is_right_punct
and not word.is_bracket and not word.is_quote
and not word.is_currency
# not word.is_punct and
# word.tag_ not in ["SYM", "HYPH"] and
# word.lemma_ != "-PRON-"
))
])
if sentences == "": continue
# if withdot == True: sentences = sentences + "."
if summary: sentences = ' <s> ' + sentences + " </s> "
sentences = sentences + "\n"
if not withdot: sentences = sentences.replace(" . ", " ")
text_lines.append(sentences)
text_lines = "".join(text_lines)
return text_lines
def get_Lemm_NLTK(text, remove_stopwords=False, summary=False):
text = text.lower()
for k, v in contractions.items():
if k in text:
text = text.replace(k, v)
for k in html_escape_table:
if k in text:
text = text.replace(k, "")
lines = nltk.sent_tokenize(text)
# Separate out reviewText and summary sentences
text_lines = []
for idx, line in enumerate(lines):
if line == "":
continue # empty line
else:
word_tokens = word_tokenize(line)
new_word_tokens = []
for w in word_tokens:
new_word_tokens.append(w)
word_tokens = new_word_tokens
word_tokens = pos_tag(word_tokens) # 获取单词词性
wnl = WordNetLemmatizer()
# print(line)
filtered_sentence = []
for w in word_tokens:
word, tag = w[0], w[1]
# if word not in stop_words:
wordnet_pos = get_wordnet_pos(tag) # or wordnet.NOUN
if wordnet_pos != None:
lemmatize_word = wnl.lemmatize(word,
pos=wordnet_pos) # 词形还原
filtered_sentence.append(lemmatize_word)
else:
filtered_sentence.append(word)
line = " ".join(filtered_sentence)
if summary:
text_lines.append(" <s> " + line + " </s> ")
else:
text_lines.append(line)
# Make reviewText into a single string
text = ' '.join(text_lines)
# text = " ".join(sentences)
text = remove_word2(text)
# Optionally, remove stop words
text = text.split()
stops = set(stopwords.words("english"))
if remove_stopwords:
text = [
w for w in text
if not (w in (stops or stpwords_list1 or stpwords_list2))
]
else:
text = [
w for w in text if not (w in (html_escape_table or stpwords_list2))
]
text = " ".join(text)
text = text + "."
return text
def lemm_sent_process(text,
remove_stopwords=False,
summary=False,
mode="nltk",
withdot=False):
text = text.lower()
stops = set(stopwords.words("english"))
for k, v in contractions.items():
if k in text:
text = text.replace(k, v)
for k in html_escape_table:
if k in text:
text = text.replace(k, "")
lines = nltk.sent_tokenize(text)
# Separate out reviewText and summary sentences
text_lines = []
for idx, line in enumerate(lines):
if line == "":
continue # empty line
else:
line = remove_word2(remove_word(line))
# nltk
if mode == "nltk":
# ------------------------------------------------------------------
word_tokens = word_tokenize(line)
new_word_tokens = []
for w in word_tokens:
new_word_tokens.append(w)
word_tokens = new_word_tokens
word_tokens = pos_tag(word_tokens) # 获取单词词性
wnl = WordNetLemmatizer()
# print(line)
filtered_sentence = []
for w in word_tokens:
word, tag = w[0], w[1]
if word in list(html_escape_table + stpwords_list2):
continue
if remove_stopwords:
if word in list(
list(stops) + list(stpwords_list1) +
list(stpwords_list3)):
continue
# if word not in stop_words:
wordnet_pos = get_wordnet_pos(tag) # or wordnet.NOUN
if wordnet_pos != None:
lemmatize_word = wnl.lemmatize(word,
pos=wordnet_pos) # 词形还原
filtered_sentence.append(lemmatize_word)
else:
filtered_sentence.append(word)
# filtered_sentence = [sent for sent in filtered_sentence if sent != "" ]
line = " ".join(filtered_sentence)
# line = remove_word2(remove_word(line))
if line == "": continue
if summary:
text_lines.append(" <s> " + line + " </s> ")
else:
if withdot == True:
text_lines.append(line + ".\n")
else:
text_lines.append(line + "\n")
# --------------------------------------------------------------------------------
# spacy
elif mode == "spacy":
# line = remove_word2(remove_word(line))
doc = nlp(line)
# 通過設置doc.is_parsed = True來欺騙spaCy忽略它，即通過讓它相信分配了依賴關係解析並以這種方式應用了句子邊界
# doc.is_parsed = True
sentences = " ".join([
word.lemma_ if word.pos_.startswith('N')
or word.pos_.startswith('J') or word.pos_.startswith('V')
or word.pos_.startswith('R') else word.orth_
for word in doc
if ((not word.is_space and not word.is_bracket
and not word.is_digit and not word.is_left_punct
and not word.is_right_punct and not word.is_bracket
and not word.is_quote and not word.is_currency
and not word.is_punct and (word.lemma_ not in list(
html_escape_table + stpwords_list2))
and not (word.is_stop and remove_stopwords)
and word.tag_ != "SYM" and word.tag_ != "HYPH"
and word.lemma_ != "-PRON-" and
(word.lemma_ not in stpwords_list3))
or word.lemma_ in ["the", "not"])
])
if sentences == "": continue
if withdot == True:
text_lines.append(sentences + ".\n")
else:
text_lines.append(sentences + "\n")
# --------------------------------------------------------------------------------
text_lines = "".join(text_lines)
return text_lines
def get_pos_sequence(text,
remove_stopwords=False,
summary=False,
mode="spacy",
withdot=False):
text = text.lower()
stops = set(stopwords.words("english"))
for k, v in contractions.items():
if k in text:
text = text.replace(k, v)
for k in html_escape_table:
if k in text:
text = text.replace(k, "")
lines = nltk.sent_tokenize(text)
# Separate out reviewText and summary sentences
text_lines = []
for idx, line in enumerate(lines):
if line == "":
continue # empty line
else:
line = remove_word2(remove_word(line))
# nltk
if mode == "nltk":
# ------------------------------------------------------------------
word_tokens = word_tokenize(line)
new_word_tokens = []
for w in word_tokens:
new_word_tokens.append(w)
word_tokens = new_word_tokens
word_tokens = pos_tag(word_tokens) # 获取单词词性
wnl = WordNetLemmatizer()
# print(line)
filtered_sentence = []
for w in word_tokens:
word, tag = w[0], w[1]
if word in list(html_escape_table + stpwords_list2):
continue
if remove_stopwords:
if word in list(list(stops) + list(stpwords_list1)):
continue
# if word not in stop_words:
wordnet_pos = get_wordnet_pos(tag) # or wordnet.NOUN
if wordnet_pos != None:
lemmatize_word = wnl.lemmatize(word,
pos=wordnet_pos) # 词形还原
filtered_sentence.append(lemmatize_word)
else:
filtered_sentence.append(word)
# filtered_sentence = [sent for sent in filtered_sentence if sent != "" ]
line = " ".join(filtered_sentence)
line = remove_word2(remove_word(line))
if line == "": continue
if summary:
text_lines.append(" <s> " + line + " </s> ")
else:
if withdot == True:
text_lines.append(line + ".\n")
else:
text_lines.append(line + "\n")
# --------------------------------------------------------------------------------
# spacy
elif mode == "spacy":
doc = nlp(line)
# 通過設置doc.is_parsed = True來欺騙spaCy忽略它，即通過讓它相信分配了依賴關係解析並以這種方式應用了句子邊界
# doc.is_parsed = True
sent = [
(word.tag_, word.orth_) if
((word.tag_.startswith('N') or word.tag_.startswith('J')
or word.tag_.startswith('V')
or word.tag_.startswith('R'))) else
(word.tag_, word.orth_) for word in doc
# if((not word.is_space and
# not word.is_bracket and
# not word.is_digit and
# not word.is_left_punct and
# not word.is_right_punct and
# not word.is_bracket and
# not word.is_quote and
# not word.is_currency and
# not word.is_punct and
# not (word.is_stop and remove_stopwords) and
# word.tag_ != "SYM" and word.tag_ != "HYPH" and word.tag_ != "HYPH" and word.ent_type_ == ""
# and word.ent_type_ not in ["PERSON", "NORP", "FAC", "ORG", "GPE", "LOC", "PRODUCT",
# "EVENT",
# "WORK_OF_ART", "LAW", "LANGUAGE", "DATE", "TIME",
# "PERCENT",
# "MONEY", "QUANTITY", "ORDINAL", "CARDINAL"]
# and word.lemma_ != "-PRON-" and (word.lemma_ not in stpwords_list2)))
]
if sent == "": continue
text_lines.append(sent)
if len(text_lines) != 0:
pos_lines = []
for line in text_lines:
pos_sent = ' '.join([word + '/' + pos for pos, word in line])
pos_lines.append(pos_sent)
return text_lines, pos_lines
else:
return text_lines, None
import re
import spacy
# gpu = spacy.prefer_gpu()
# print('GPU:', gpu)
from spacy.matcher import Matcher
import en_core_web_sm
# nlp = en_core_web_sm.load(disable = ['ner', 'tagger','parser'])
nlp = en_core_web_sm.load()
from nltk.corpus import stopwords
# from stopwords import *
stops = set(stopwords.words("english"))
stpwords_list3 = [
f.replace("\n", "") for f in open(
"data_util/stopwords.txt", "r", encoding="utf-8").readlines()
]
stpwords_list3.remove("not")
total_stopwords = list(html_escape_table + stpwords_list2) + list(
list(stops) + list(stpwords_list1) + list(stpwords_list3))
total_stopwords = set(total_stopwords)
total_stopwords.remove("the")
total_stopwords.remove("not")
ent_list = [
'PERSON', 'NORP', 'FAC', 'ORG', 'GPE', 'LOC', 'PRODUCT', 'EVENT',
'WORK_OF_ART', 'LAW', 'LANGUAGE', 'DATE', 'TIME', 'PERCENT', 'MONEY',
'QUANTITY', 'ORDINAL', 'CARDINAL'