def test_replace_emojis():
in_outs = [
("ugh, it's raining *again* ☔", "ugh, it's raining *again* _EMOJI_"),
("✌ tests are passing ✌", "_EMOJI_ tests are passing _EMOJI_"),
]
for in_, out_ in in_outs:
assert preprocessing.replace_emojis(in_) == out_
def preprocess_siamese(doc):
# pre-process data
doc = tp.normalize.normalize_unicode(doc)
doc = tp.normalize_whitespace(doc)
doc = tp.normalize_quotation_marks(doc)
doc = tp.replace_emails(doc, replace_with="<EMAIL>")
doc = tp.replace_urls(doc, replace_with="<URL>")
doc = tp.replace_hashtags(doc, replace_with="<HASHTAG>")
doc = tp.replace_emojis(doc, replace_with="<EMOJI>")
doc = tp.replace_phone_numbers(doc, replace_with="<PHONE>")
# apply spaCy to tokenize doc
doc = nlp_token(doc)
# build new sentences for pre-processed doc
doc_new = []
for sent in doc.sents:
sent_new = ""
for token in sent:
token = token.text
token = token.replace("\n", "")
token = token.replace("\t", "")
token = token.strip()
sent_new += token + " "
doc_new.append(sent_new[:-1])
return doc_new
def cleanTweet(raw_data):
data = [txt.replace_urls(x, "") for x in raw_data]
data = [txt.replace_emails(x, "") for x in data]
data = [txt.replace_emojis(x, "") for x in data]
data = [txt.replace_user_handles(x, "") for x in data]
data = [txt.replace_phone_numbers(x, "") for x in data]
data = [txt.replace_numbers(x, "") for x in data]
data = [txt.replace_currency_symbols(x, "") for x in data]
data = [txt.replace_hashtags(x, "") for x in data]
return data
def checkLenTweet(raw_data):
data = txt.replace_urls(raw_data, "")
data = txt.replace_emails(data, "")
data = txt.replace_emojis(data, "")
data = txt.replace_user_handles(data, "")
data = txt.replace_phone_numbers(data, "")
data = txt.replace_numbers(data, "")
data = txt.replace_currency_symbols(data, "")
data = txt.replace_hashtags(data, "")
if len(data) < 28: #Lunghezza media di un tweet
return False
else:
return True
def preprocess_sentence(sent, lower=True):
"""Pre-process a sentence ( via ``textacy.preprocess` module ).
Args:
sent (str): text.
lower (bool): whether to return a lowercase string.
Returns:
str
"""
# normalize unicode
sent = preprocessing.normalize_unicode(sent)
# deaccent
sent = preprocessing.remove_accents(sent)
# replace newline chars
sent = re.sub("\n|\r", " ", sent)
# unpack contractions
sent = contractions.fix(sent)
# replace emoji symbols
sent = preprocessing.replace_emojis(sent)
# replace hashtags
sent = preprocessing.replace_hashtags(sent)
# replace user handles
sent = preprocessing.replace_user_handles(sent)
# replace currency symbols
sent = preprocessing.replace_currency_symbols(sent)
# replace emails
sent = preprocessing.replace_emails(sent)
# replace URLs
sent = preprocessing.replace_urls(sent)
# remove punctuation
sent = preprocessing.remove_punctuation(sent)
# normalize whitespace
sent = preprocessing.normalize_whitespace(sent)
if lower:
sent = sent.lower()
return sent
def textacy_preprocess(sentence):
"""Preprocess text."""
sentence = preprocessing.normalize_hyphenated_words(sentence)
sentence = preprocessing.normalize_quotation_marks(sentence)
#sentence = preprocessing.normalize_repeating_chars(sentence)
sentence = preprocessing.normalize_unicode(sentence)
sentence = preprocessing.normalize_whitespace(sentence)
sentence = preprocessing.remove_accents(sentence)
sentence = preprocessing.remove_punctuation(sentence)
sentence = preprocessing.replace_currency_symbols(sentence)
sentence = preprocessing.replace_emails(sentence)
sentence = preprocessing.replace_emojis(sentence)
sentence = preprocessing.replace_hashtags(sentence)
sentence = preprocessing.replace_numbers(sentence)
sentence = preprocessing.replace_phone_numbers(sentence)
sentence = preprocessing.replace_urls(sentence)
sentence = preprocessing.replace_user_handles(sentence)
return sentence
def preprocess_text(text,
char_count_filter=True,
stopwords=None,
min_len=2,
max_len=15):
"""
Pre-processing steps prior to spaCy nlp pipeline. Optional filtering of
tokens based on character length.
Parameters
----------
text : str
char_count_filter : bool
stopwords : iterable, None
min_len : int
max_len : int
Returns
-------
text : str
pre-processed text
"""
# 1) convert to lower case for robust stop-word recognition
text = text.lower()
# 2) normalise
text = preprocessing.normalize_quotation_marks(text)
# text = preprocessing.normalize_repeating_chars(text)
text = preprocessing.normalize_hyphenated_words(text)
text = preprocessing.normalize_whitespace(text)
# 3) replace
text = preprocessing.replace_currency_symbols(text)
text = preprocessing.replace_emails(text)
text = preprocessing.replace_emojis(text)
text = preprocessing.replace_hashtags(text)
text = preprocessing.replace_numbers(text)
text = preprocessing.replace_phone_numbers(text)
text = preprocessing.replace_urls(text)
text = preprocessing.replace_user_handles(text)
# 4) remove
text = preprocessing.remove_accents(text)
text = preprocessing.remove_punctuation(text)
text = re.sub("[^A-Za-z0-9]+", " ", text) # keep text and numbers
# 5) optionally remove tokens based on length
if char_count_filter & (stopwords is not None):
# filter based on token length
tokens = gensim.utils.simple_preprocess(doc=text,
min_len=min_len,
max_len=max_len)
# filter case-specific words
tokens = [token for token in tokens if token not in stopwords]
# convert processed list of tokens back to one string
text = " ".join(tokens)
else:
raise NotImplementedError("Not implemented.")
return text
