def tokenize_okt_noscreen(df):
okt = Twitter()
okt.add_dictionary(call_userword(), 'Noun')
stopwords = load_wordset('./tokenizer/korean_stopword.txt')
#stopwords = stopwords | load_wordset('./tokenizer/korean_screen.txt')
stopwords = list(stopwords)
df['content_token'] = df.progress_apply(lambda x: [t[0] for t in okt.pos(
x['content'], stem=True) if t[1] in ['Noun', 'Verb', 'Adjective'] and t[0] not in stopwords and len(t[0]) != 1], axis=1)
df['title_token'] = df.progress_apply(lambda x: [t[0] for t in okt.pos(
x['title'], stem=True) if t[1] in ['Noun', 'Verb', 'Adjective'] and t[0] not in stopwords and len(t[0]) != 1], axis=1)
return df
class RawTaggerReader:
def __init__(self, filepath, tagger=None):
if tagger:
self.tagger = tagger
else:
self.tagger = Twitter()
self.filepath = filepath
self.rgxSplitter = re.compile('([.!?:](?:["\']|(?![0-9])))')
def __iter__(self):
for line in open(self.filepath, encoding='utf-8'):
ch = self.rgxSplitter.split(line)
for s in map(lambda a, b: a + b, ch[::2], ch[1::2]):
if not s: continue
yield self.tagger.pos(s)
class RawTagger:
def __init__(self, textIter, tagger=None):
# 전처리 1 (형태소 분석 단어 추가) -> 형태소 분석기가 단어로 인식 못하는 단어들을 추가
f = open('형태소 보완.txt')
dd = f.read()
a = dd.split('\n')
if tagger:
self.tagger = tagger
else:
self.tagger = Twitter()
self.tagger.add_dictionary(a, 'Noun')
if type(textIter) == str:
self.textIter = textIter.split('\n')
else:
self.textIter = textIter
self.rgxSplitter = re.compile('([.!?:](?:["\']|(?![0-9])))')
def __iter__(self):
for line in self.textIter:
ch = self.rgxSplitter.split(line)
for s in map(lambda a, b: a + b, ch[::2], ch[1::2]):
if not s: continue
yield self.tagger.pos(s)
class Social_analysis():
non_bmp_map = dict.fromkeys(range(0x10000, sys.maxunicode + 1), 0xfffd)
def __init__(self):
self.twitter = Twitter()
def pickle_to_table(self, filename):
with open(filename, 'rb') as f:
data = pickle.load(f)
data = data[1:]
for idx, i in enumerate(data):
data[idx][2] = i[2].replace('#', ' ').translate(self.non_bmp_map)
data[idx][3] = '/'.join(i[3])
data[idx][4] = '/'.join(i[4])
self.raw_data = np.array(data)
def hashtags_split(self, hashtags):
hashtags_split = []
for i in hashtags:
hashtags_split.append(i.split('/'))
hashtags_list = []
for i in hashtags_split:
temp = []
for j in i:
if self.isHangul(j):
t_hashtags = j.translate(self.non_bmp_map)
temp.append(t_hashtags)
hashtags_list.append(temp)
self.hashtags_list = hashtags_list
return hashtags_list
def add_keyword_dic(self, keyword_list, tag='Noun'):
for i in keyword_list:
if type(i) == tuple:
self.twitter.add_dictionary(i[0], i[1])
else:
self.twitter.add_dictionary(i, tag)
def morph_pos(self, text_list, exception_list=['맛', '밥', '물', '몸']):
morph_list = []
noun_list = []
adj_list = []
verb_list = []
for j in text_list:
parsed = self.twitter.pos(j)
temp = []
n_temp = []
adj_temp = []
verb_temp = []
for i in parsed:
if self.isHangul(i[0]):
if ((len(i[0]) > 1) or (i[0] in exception_list)):
temp.append(i)
if i[1] == 'Noun':
n_temp.append(i[0])
elif i[1] == 'Verb':
adj_temp.append(i[0])
elif i[1] == 'Adjective':
verb_temp.append(i[0])
else:
print('{} 제외'.format(i[0]))
else:
print('{} 한글이 아님.'.format(i[0]))
morph_list.append(temp)
noun_list.append(n_temp)
adj_list.append(adj_temp)
verb_list.append(verb_temp)
nav_list = noun_list + adj_list + verb_list
return morph_list, nav_list, noun_list, adj_list, verb_list
def merge_list(self, tokenized_list):
return [j for i in tokenized_list for j in i]
def join_list(self, tokenized_list):
joined_list = []
for idx, i in enumerate(tokenized_list):
joined_list.append(" ".join(i))
return joined_list
def split_list(self, untokenized_list):
hashtag_splited = []
for idx, i in enumerate(untokenized):
hashtag_splited.append(i.split('/'))
return hastag_splited
def word_substitute(self, dataset, sublist):
dataset = copy.deepcopy(dataset)
sub_book = dict()
for i in sublist:
for j in i['sub_words']:
sub_book[j] = i['main']
gc.collect()
for n, i in enumerate(dataset):
dataset[n] = [sub_book.get(item, item) for item in i]
del sub_book
gc.collect()
return dataset
def word_delete(self, dataset, del_list):
dataset = copy.deepcopy(dataset)
for n, line in enumerate(dataset):
dataset[n] = [i for i in line if i not in del_list]
return dataset
def isHangul(self, text):
encText = text
hanCount = len(re.findall(u'[\u3130-\u318F\uAC00-\uD7A3]+', encText))
return hanCount > 0
def token(self, title, ccontent, creplies):
memory = psutil.Process(os.getpid())
T_OR_title = []
T_title = []
T_OR_ccontent = []
T_ccontent = []
T_OR_creplies = []
T_creplies = []
twitter = Okt() # 트위터 형태소 사전을 사용하기 위해 초기화
twitter.add_dictionary('백래시', 'Noun')
twitter.add_dictionary('문재앙', 'Noun')
#### 타이틀 토큰화
#print('1')
for i in range(len(title)):
a = twitter.pos(title[i])
b = []
#print('title[i]',i,title[i])
for j in range(len(a)):
if a[j][1] != 'Punctuation': # 오류로 'Punctuation'에 해당하는 튜플 제거
b.append(a[j])
#print('3',j)
T_OR_title.append(b)
T_title.append(twitter.morphs(title[i]))
#### ccontent 토큰화
try:
c = twitter.pos(str(ccontent[i]))
d = []
# print('ccontent[i]',i, ccontent[i])
for w in range(len(c)):
if c[w][1] != 'Punctuation': # 오류로 'Punctuation'에 해당하는 튜플 제거
d.append(c[w])
#print('4',w)
T_OR_ccontent.append(d)
T_ccontent.append(twitter.morphs(str(ccontent[i])))
except RuntimeError as e:
T_OR_ccontent.append('')
T_ccontent.append(twitter.morphs(''))
### 댓글 토큰화
#print('creplies[i]',i,creplies[i])
if type(creplies[i]) == str: # string형 댓글 토큰화
a = [creplies[i]] # string을 리스트로 변경
e = twitter.pos(str(a))
f = []
for u in range(len(e)):
if e[u][1] != 'Punctuation':
f.append(e[u])
elif e[u][1] != 'KoreanParticle':
f.append(e[u])
else:
break
#print('5',u)
T_OR_creplies.append(f)
T_OR_creplies.append(twitter.pos(str(a)))
T_creplies.append(twitter.morphs(str(a)))
else:
temp = []
temp2 = []
x = []
for n in range(len(creplies[i])): ### 리스트로 반환되는 댓글
h = twitter.pos(creplies[i][n])
#print('6',n)
for z in range(len(h)):
if h[z][1] != 'Punctuation':
x.append(h[z])
elif h[z][1] != 'KoreanParticle':
x.append(h[z])
else:
break
# print('7',z)
# print('8',)
#print('h',z,h)
temp.append(x)
temp2.append(twitter.morphs(creplies[i][n]))
T_OR_creplies.append(temp)
T_creplies.append(temp2)
return T_OR_title, T_title, T_OR_ccontent, T_ccontent, T_OR_creplies, T_creplies
class Social_analysis():
non_bmp_map = dict.fromkeys(range(0x10000, sys.maxunicode + 1), 0xfffd)
def __init__(self):
self.twitter = Twitter()
def pickle_to_table(self, filename):
with open(filename, 'rb') as f:
data = pickle.load(f)
data = data[1:]
for idx, i in enumerate(data):
data[idx][2] = i[2].replace('#', ' ').translate(self.non_bmp_map)
data[idx][3] = '/'.join(i[3])
data[idx][4] = '/'.join(i[4])
self.raw_data = np.array(data)
def DB_to_table(self, DBname='intake', keyword='intake'):
self.query = \
"""
SELECT keyword, created_at, post_name, main_text, current_url FROM NaverBlogReview WHERE keyword = '{}'
""".format(keyword)
conn = pymssql.connect(
"intakedb.c63elkxbiwfc.us-east-2.rds.amazonaws.com:1433", "gh",
"ghintake", DBname)
df = pdsql.read_sql_query(self.query, con=conn)
# df['main_text'] = df.main_text.apply(lambda x: x.replace('#',' ').translate(self.non_bmp_map))
# df['created_at'] = df.created_at.apply(lambda x: x.strftime("%Y-%m-%d %H:%M:%S"))
conn.close()
self.raw_data = df.as_matrix()
# def hashtags_split(self, hashtags):
# hashtags_split = []
# for i in hashtags:
# hashtags_split.append(i.split('/'))
#
# hashtags_list = []
#
# for i in hashtags_split:
# temp = []
# for j in i:
# if self.isHangul(j):
# t_hashtags = j.translate(self.non_bmp_map)
# temp.append(t_hashtags)
# hashtags_list.append(temp)
# self.hashtags_list = hashtags_list
#
# return hashtags_list
def add_keyword_dic(self, keyword_list, tag='Noun'):
for i in keyword_list:
if type(i) == tuple:
self.twitter.add_dictionary(i[0], i[1])
else:
self.twitter.add_dictionary(i, tag)
def morph_pos(self, text_list, exception_list=['맛', '밥', '물', '몸']):
morph_list = []
noun_list = []
adj_list = []
verb_list = []
for j in text_list:
parsed = self.twitter.pos(j)
temp = []
n_temp = []
adj_temp = []
verb_temp = []
for i in parsed:
if self.isHangul(i[0]):
if ((len(i[0]) > 1) or (i[0] in exception_list)):
temp.append(i)
if i[1] == 'Noun':
n_temp.append(i[0])
elif i[1] == 'Verb':
adj_temp.append(i[0])
elif i[1] == 'Adjective':
verb_temp.append(i[0])
else:
print('{} 제외'.format(i[0]))
else:
print('{} 한글이 아님.'.format(i[0]))
morph_list.append(temp)
noun_list.append(n_temp)
adj_list.append(adj_temp)
verb_list.append(verb_temp)
nav_list = noun_list + adj_list + verb_list
return morph_list, nav_list, noun_list, adj_list, verb_list
def merge_list(self, tokenized_list):
return [j for i in tokenized_list for j in i]
def join_list(self, tokenized_list):
joined_list = []
for idx, i in enumerate(tokenized_list):
joined_list.append(" ".join(i))
return joined_list
def split_list(self, untokenized_list):
hashtag_splited = []
for idx, i in enumerate(untokenized):
hashtag_splited.append(i.split('/'))
return hastag_splited
def word_substitute(self, dataset, sublist):
dataset = copy.deepcopy(dataset)
sub_book = dict()
for i in sublist:
for j in i['sub_words']:
sub_book[j] = i['main']
gc.collect()
for n, i in enumerate(dataset):
dataset[n] = [sub_book.get(item, item) for item in i]
del sub_book
gc.collect()
return dataset
def word_delete(self, dataset, del_list):
dataset = copy.deepcopy(dataset)
for n, line in enumerate(dataset):
dataset[n] = [i for i in line if i not in del_list]
return dataset
def isHangul(self, text):
encText = text
hanCount = len(re.findall(u'[\u3130-\u318F\uAC00-\uD7A3]+', encText))
return hanCount > 0
from matplotlib import font_manager, rc
from PIL import Image
import numpy as np
font_name = font_manager.FontProperties(
fname="c:/Windows/Fonts/malgun.ttf").get_name()
rc('font', family=font_name)
tager = Twitter()
for add in add_noun:
tager.add_dictionary(add, 'Noun')
text = open(keyword + '.txt', encoding='utf-8-sig').read()
tokens = tager.pos(text)
wordlist = []
for word in tokens:
if word[1] in ['Noun']:
if word[0] not in no_word:
wordlist.append(word[0])
words = nltk.Text(wordlist, name=keyword)
top_words = words.vocab().most_common(1000)
words_dic = dict(top_words)
mask = np.array(Image.open("shape.png"))
wordcloud = WordCloud(
font_path='c:/Windows/Fonts/malgun.ttf',
mask=mask,
filtered_content = re.sub(r'[^\.\?\!\w\d\s]', '', filtered_content)
print(filtered_content)
from ckonlpy.tag import Twitter
twitter = Twitter()
# 가상통화, 아시아경제 등 미등록 단어를 사전에 추가로 등록
twitter.add_dictionary('가상통화', 'Noun')
twitter.add_dictionary('아시아경제', 'Noun')
twitter.add_dictionary('한동안', 'Noun')
twitter.add_dictionary('블리클리', 'Noun')
twitter.add_dictionary('공동창립자', 'Noun')
twitter.add_dictionary('부크바', 'Noun')
# 형태소 분석
twitter_morphs = twitter.pos(filtered_content)
# 명사만 추출하기
Noun_words = []
for word, pos in twitter_morphs:
if pos == 'Noun':
Noun_words.append(word)
print(Noun_words)
# 불용어 제거를 위한 별도의 사전 구축
# 본문과 상관 없는 아시아 경제, 기자 이름, 기자 단어 제거
# count 했을 때 빈도수가 높지만 본문 주요 내용과 관련이 없는 못, 것, 수, 까지 단어 제거
stopwords = ['아시아경제', '김철현', '기자', '못', '것', '수', '까지']
# unique 하게 명사 추려내기
unique_noun_words = set(Noun_words)
# 트위터 tokenizer loading 및 업로드
twitter = Twitter()
twitter.add_dictionary(brand_set, 'Noun')
# 원래 tokenizer 업로드
ori_twitter = original_Twitter()
#### emoji2text
preprocessed_text = emoji2text(review_DB.text)
#### upper & 제품 매핑
preprocessed_text = pd.Series(preprocessed_text).apply(lambda x: x.upper())
bow = preprocessed_text.apply(lambda x: '◈'.join([
token + '╹' + pos if pos in ['Noun'] else ori_twitter.
pos(token, stem=True, norm=True)[0][0] + '╹' + ori_twitter.pos(
token, stem=True, norm=True)[0][1] for token, pos in twitter.pos(x)
]))
### token normalizing
p_normal = {
'BB╹Alpha': '비비크림╹Noun',
'CC╹Alpha': '씨씨크림╹Noun',
'비비╹Noun': '비비크림╹Noun',
'씨씨╹Noun': '씨씨크림╹Noun',
'파데╹Noun': '파운데이션╹Noun',
'쟂빛╹Noun': '잿빛╹Noun',
'비씨╹Noun': '비씨데이션╹Noun'
}
bow = bow.apply(lambda x: '◈'.join([
p_normal[token] if token in p_normal.keys() else token
for token in x.split('◈')
]))
class Social_analysis():
non_bmp_map = dict.fromkeys(range(0x10000, sys.maxunicode + 1), 0xfffd)
def __init__(self):
self.twitter = Twitter()
def DB_to_table(self, DBname='intake', keyword='intake'):
import pymssql
import pandas.io.sql as pdsql
import pandas as pd
self.query = \
"""
SELECT user_id, created_at, main_text, hashtags, comments, likes, current_url FROM instaPost WHERE keyword = '{}'
""".format(keyword)
conn = pymssql.connect(
"intakedb.c63elkxbiwfc.us-east-2.rds.amazonaws.com:1433", "gh",
"ghintake", DBname)
df = pdsql.read_sql_query(self.query, con=conn)
# df['main_text'] = df.main_text.apply(lambda x: x.replace('#',' ').translate(self.non_bmp_map))
# df['created_at'] = df.created_at.apply(lambda x: x.strftime("%Y-%m-%d %H:%M:%S"))
conn.close()
self.raw_data = df.as_matrix()
def pickle_to_table(self, filename):
with open(filename, 'rb') as f:
data = pickle.load(f)
data = data[1:]
for idx, i in enumerate(data):
data[idx][2] = i[2].replace('#', ' ').translate(self.non_bmp_map)
data[idx][3] = '/'.join(i[3])
data[idx][4] = '/'.join(i[4])
self.raw_data = np.array(data)
def hashtags_split(self, hashtags):
hashtags_split = []
for i in hashtags:
hashtags_split.append(i.split('/'))
hashtags_list = []
for i in hashtags_split:
temp = []
for j in i:
if self.isHangul(j):
t_hashtags = j.translate(self.non_bmp_map)
temp.append(t_hashtags)
hashtags_list.append(temp)
self.hashtags_list = hashtags_list
return hashtags_list
def add_keyword_dic(self, keyword_list, tag='Noun'):
for i in keyword_list:
if type(i) == tuple:
self.twitter.add_dictionary(i[0], i[1])
else:
self.twitter.add_dictionary(i, tag)
def morph_pos(self, text_list, exception_list=['맛', '밥', '물', '몸']):
morph_list = []
noun_list = []
adj_list = []
verb_list = []
nav_list = []
for j in text_list:
parsed = self.twitter.pos(j)
temp = []
n_temp = []
adj_temp = []
verb_temp = []
nav_temp = []
for i in parsed:
if self.isHangul(i[0]):
if ((len(i[0]) > 1) or (i[0] in exception_list)):
temp.append(i)
if i[1] == 'Noun':
n_temp.append(i[0])
nav_temp.append(i[0])
elif i[1] == 'Verb':
adj_temp.append(i[0])
nav_temp.append(i[0])
elif i[1] == 'Adjective':
verb_temp.append(i[0])
nav_temp.append(i[0])
else:
print('{} 제외'.format(i[0]))
else:
print('{} 한글이 아님.'.format(i[0]))
morph_list.append(temp)
noun_list.append(n_temp)
adj_list.append(adj_temp)
verb_list.append(verb_temp)
nav_list.append(nav_temp)
return morph_list, nav_list, noun_list, adj_list, verb_list
def pos_extractor(self, parsed):
noun_list = []
adj_list = []
verb_list = []
nav_list = []
for i in parsed:
n_temp = []
adj_temp = []
verb_temp = []
nav_temp = []
if self.isHangul(i[0]):
if ((len(i[0]) > 1) or (i[0] in exception_list)):
if i[1] == 'Noun':
n_temp.append(i[0])
nav_temp.append(i[0])
elif i[1] == 'Verb':
adj_temp.append(i[0])
nav_temp.append(i[0])
elif i[1] == 'Adjective':
verb_temp.append(i[0])
nav_temp.append(i[0])
else:
print('{} 제외'.format(i[0]))
else:
print('{} 한글이 아님.'.format(i[0]))
noun_list.append(n_temp)
adj_list.append(adj_temp)
verb_list.append(verb_temp)
nav_list.append(nav_temp)
return nav_list, noun_list, adj_list, verb_list
def merge_list(self, tokenized_list):
return [j for i in tokenized_list for j in i]
def join_list(self, tokenized_list):
joined_list = []
for idx, i in enumerate(tokenized_list):
joined_list.append(" ".join(i))
return joined_list
def split_list(self, untokenized_list):
hashtag_splited = []
for idx, i in enumerate(untokenized):
hashtag_splited.append(i.split('/'))
return hastag_splited
def join_underbar(self, morph_list):
all_list = []
post_list = []
for i in morph_list:
for j in i:
post_list.append(j[0] + '_' + j[1])
all_list.append([(' , ').join(post_list)])
post_list = []
all_list = np.array(all_list)
return all_list
def word_substitute(self, dataset, sublist):
dataset = copy.deepcopy(dataset)
sub_book = dict()
for i in sublist:
for j in i['sub_words']:
sub_book[j] = i['main']
gc.collect()
for n, i in enumerate(dataset):
dataset[n] = [sub_book.get(item, item) for item in i]
del sub_book
gc.collect()
return dataset
def word_delete(self, dataset, del_list):
dataset = copy.deepcopy(dataset)
for n, line in enumerate(dataset):
dataset[n] = [i for i in line if i not in del_list]
return dataset
def isHangul(self, text):
encText = text
hanCount = len(re.findall(u'[\u3130-\u318F\uAC00-\uD7A3]+', encText))
return hanCount > 0
def convert_list(self, *tokenized_list):
input_length = len(tokenized_list)
lists = [[] for i in range(input_length)]
for idx, li in enumerate(tokenized_list):
for j in li:
lists[idx].append(['/'.join(j)])
converted_array = np.array(lists[0])
for idx in range(input_length):
try:
converted_array = np.concatenate(
(converted_array, lists[idx + 1]), axis=1)
except Exception as e:
print(e, '끝')
return converted_array
def make_df(self, converted_array):
df = pd.DataFrame(np.hstack(
(intake.raw_data[:, :3], converted_array, intake.raw_data[:, 3:])),
index=None)
return df
# 키워드 리스트 중 하나라도 있는 경우
def word_check_or(self, text, keywords):
if any(word in text for word in keywords): return True
else: return False
# 키워드 리스트에 있는 단어가 모두 있는 경우
def word_check_and(self, text, keywords):
if all(word in text for word in keywords): return True
else: return False
def word_check(self,
method='and',
keywords=[],
df=None,
column_name=None,
filter_TF=True):
if method == 'and':
df['flag'] = df[column_name].apply(
lambda x: self.word_check_and(x, keywords))
return df[df.flag == filter_TF]
if method == 'or':
df['flag'] = df[column_name].apply(
lambda x: self.word_check_or(x, keywords))
return df[df.flag == filter_TF]
from konlpy.tag import Komoran
from konlpy.tag import Okt
from ckonlpy.tag import Twitter
okt = Okt()
twitter = Twitter()
sentence = 'IBK기업은행 '
sentences = '소은지국민은행계좌로30만원이체해줘'
komoran = Komoran()
twitter.add_dictionary('이체해줘', 'Noun')
twitter.add_dictionary('KB 국민은행', 'Noun')
komoran = Komoran(userdic="C:/Users/ADMIN/Desktop/dic.txt")
print(twitter.pos(sentence, stem=True))
print(twitter.pos(sentences, stem=True))
print(komoran.pos(sentence))
print(komoran.pos(sentences))
arr = komoran.pos(sentence)
for word, tag in arr:
if (tag == 'VV'): print("|||||||")
print(word, tag)
if (tag == 'JKO' or tag == 'JKB' or tag == 'JKS'): print("|||||||")
brr = komoran.pos(sentences)
for word, tag in brr:
if (tag == 'VV' or tag == 'XSV'):
