-
Notifications
You must be signed in to change notification settings - Fork 0
/
preprocess_covidbert.py
182 lines (162 loc) · 6.43 KB
/
preprocess_covidbert.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
## Preprocessing code from https://github.com/digitalepidemiologylab/covid-twitter-bert
import os
import re
import logging
import unicodedata
from html.parser import HTMLParser
import emoji
import unidecode
from spacy.lang.en import English
logger = logging.getLogger(__name__)
# build spacy model
def build_spacy_model():
def _avoid_sentence_boundary_on_hashtag(doc):
for token in doc[:-1]:
if token.text == '#':
doc[token.i+1].is_sent_start = False
return doc
nlp = English()
sentencizer = nlp.create_pipe("sentencizer")
nlp.add_pipe(sentencizer)
nlp.add_pipe(_avoid_sentence_boundary_on_hashtag)
return nlp
nlp = build_spacy_model()
# compile regexes
username_regex = re.compile(r'(^|[^@\w])@(\w{1,15})\b')
url_regex = re.compile(r'((www\.[^\s]+)|(https?://[^\s]+)|(http?://[^\s]+))')
control_char_regex = re.compile(r'[\r\n\t]+')
# translate table for punctuation
transl_table = dict([(ord(x), ord(y)) for x, y in zip(u"‘’´“”–-", u"'''\"\"--")])
# HTML parser
html_parser = HTMLParser()
corona = "corona virus, covid, covid 19, covid-19, covid19, covid 2019, covid2019, \
covd, coronavirus disease 2019, sars cov 2, sarscov 2, sars cov2, sarscov2, ncov, \
koronavirus, korona virus, wuhan corona virus, wuhan coronavirus, wuhancoronavirus".split(',')
def preprocess_bert(text):
"""Preprocesses tweet for BERT"""
# standardize
text = standardize_text(text)
# replace usernames/urls
text = replace_usernames(text, filler='twitteruser')
text = replace_urls(text, filler='twitterurl')
text = asciify_emojis(text)
text = standardize_punctuation(text)
text = text.lower()
text = replace_multi_occurrences(text, 'twitteruser')
text = replace_multi_occurrences(text, 'twitterurl')
text = remove_unicode_symbols(text)
# text = re.sub(r'\bcorona\b', r'coronavirus', text)
# text = re.sub(r'\bkorona\b', r'coronavirus', text)
# for wd in corona:
# text = text.replace(wd.strip(), 'coronavirus')
# for wd in ['5 g', 'fiveg', 'five g']:
# text = text.replace(wd, '5g')
return text
def remove_accented_characters(text):
text = unidecode.unidecode(text)
return text
def remove_unicode_symbols(text):
text = ''.join(ch for ch in text if unicodedata.category(ch)[0] != 'So')
return text
def replace_multi_occurrences(text, filler):
"""Replaces multiple occurrences of filler with n filler"""
# only run if we have multiple occurrences of filler
if text.count(filler) <= 1:
return text
# pad fillers with whitespace
text = text.replace(f'{filler}', f' {filler} ')
# remove introduced duplicate whitespaces
text = ' '.join(text.split())
# find indices of occurrences
indices = []
for m in re.finditer(r'{}'.format(filler), text):
index = m.start()
indices.append(index)
# collect merge list
merge_list = []
for i, index in enumerate(indices):
if i > 0 and index - old_index == len(filler) + 1:
# found two consecutive fillers
if len(merge_list) > 0 and merge_list[-1][1] == old_index:
# extend previous item
merge_list[-1][1] = index
merge_list[-1][2] += 1
else:
# create new item
merge_list.append([old_index, index, 2])
old_index = index
# merge occurrences
if len(merge_list) > 0:
new_text = ''
pos = 0
for (start, end, count) in merge_list:
new_text += text[pos:start]
new_text += f'{count} {filler}'
pos = end + len(filler)
new_text += text[pos:]
text = new_text
return text
def segment_sentences(text, args):
"""Uses spacy to segment text into sentences. Sentences which only consist of a filler will be merged with previous or following sentences"""
doc = nlp(text)
regex_fillers = r'(^\d {username}$)|^{username}$|(^\d {url}$)|^{url}$'.format(username=args.username_filler, url=args.url_filler)
num_tokens = len(doc)
sentences = [s.string.strip() for s in doc.sents]
for i, sentence in enumerate(sentences):
if re.match(regex_fillers, sentence):
if i == 0 and len(sentences) > 1:
# prepend to next sentence
sentences[i+1] = f'{sentence} {sentences[i+1]}'
elif i > 0:
# add sentence to previous
sentences[i-1] += f' {sentence}'
# remove current
del sentences[i]
return sentences, num_tokens
def asciify_emojis(text):
"""
Converts emojis into text aliases. E.g. 👍 becomes :thumbs_up:
For a full list of text aliases see: https://www.webfx.com/tools/emoji-cheat-sheet/
"""
text = emoji.demojize(text)
return text
def standardize_text(text):
"""
1) Escape HTML
2) Replaces some non-standard punctuation with standard versions.
3) Replace \r, \n and \t with white spaces
4) Removes all other control characters and the NULL byte
5) Removes duplicate white spaces
"""
# escape HTML symbols
text = html_parser.unescape(text)
# standardize punctuation
text = text.translate(transl_table)
text = text.replace('…', '...')
# replace \t, \n and \r characters by a whitespace
text = re.sub(control_char_regex, ' ', text)
# remove all remaining control characters
text = ''.join(ch for ch in text if unicodedata.category(ch)[0] != 'C')
# replace multiple spaces with single space
text = ' '.join(text.split())
return text.strip()
def standardize_punctuation(text):
return ''.join([unidecode.unidecode(t) if unicodedata.category(t)[0] == 'P' else t for t in text])
def replace_usernames(text, filler='user'):
# @<user> is a marker used internally. use filler instead
text = text.replace('@<user>', f'{filler}')
# replace other user handles by filler
text = re.sub(username_regex, filler, text)
# add spaces between, and remove double spaces again
text = text.replace(filler, f' {filler} ')
text = ' '.join(text.split())
return text
def replace_urls(text, filler='url'):
# <url> is a marker used internally. use filler instead
text = text.replace('<url>', filler)
# replace other urls by filler
text = re.sub(url_regex, filler, text)
# add spaces between, and remove double spaces again
text = text.replace(filler, f' {filler} ')
text = ' '.join(text.split())
return text