def german_semantic(text):
from nltk.corpus import stopwords
from nltk.stem.cistem import Cistem
stopwords = set(stopwords.words("german"))
liste = []
stemmer = Cistem()
wordlist = []
# clean up the text
text = "".join(text.lower())
text = text.replace('[^\w\s]', '')
text = re.sub("\s+", " ", text)
# delete stopwords
for word in text.split():
if word not in stopwords:
liste.append(word)
text = " ".join(liste)
# stemmer
for word in text.split():
word = stemmer.segment(word)[0]
wordlist.append(word)
text = " ".join(wordlist)
# sentiment
blob = TextBlobDE(text)
sentiment_polarity = blob.sentiment.polarity
sentiment_subjectivity = blob.sentiment.subjectivity
return sentiment_polarity, sentiment_subjectivity
def stem(text):
if type(text) is str:
text = tokenize(text)
stemmer = Cistem()
for index, word in enumerate(text):
text[index] = stemmer.stem(word)
return ' '.join(text)
def stem_cistem(x):
from nltk.stem.cistem import Cistem
stemmer = Cistem()
s_text = []
for word in x:
s_text.append(stemmer.stem(word))
s_text = ''.join(s_text)
return s_text
def stem(string, stemmer="porter", **kwargs):
if stemmer == "porter":
impl = PorterStemmer()
elif stemmer == "lancaster":
impl = LancasterStemmer()
elif stemmer == "regex":
regexp = kwargs['regexp']
if 'min' in kwargs:
min = kwargs['min']
else:
mins = 0
impl = RegexpStemmer(regexp=regexp, min=min)
elif stemmer == "isri":
impl = ISRIStemmer()
elif stemmer == "snowball":
if 'language' in kwargs:
language = kwargs['language']
else:
language = 'english'
impl = SnowballStemmer(language=language)
elif stemmer == "rslp":
impl = RSLPStemmer()
elif stemmer == "cistem":
if 'case_insensitive' in kwargs:
case_insensitive = kwargs['case_insensitive']
else:
case_insensitive = False
impl = Cistem(case_insensitive=case_insensitive)
else:
return string
return impl.stem(string)
def determinar_seguimiento(titulo_principal, titulo_querella):
stemmer = Cistem()
regex = r'\b\w+\b'
palabras_titulo_principal = [stemmer.stem("".join(re.findall(regex, palabra.lower()))) for palabra in titulo_principal.split(" ") if palabra not in stopwords.words('english')]
palabras_titulo_querella = [stemmer.stem("".join(re.findall(regex, palabra.lower()))) for palabra in titulo_querella.split(" ") if palabra not in stopwords.words('english')]
"""AQUI SE DEBEN INCLUIR CONSULTAS A API'S DE NLP PARA OBTENER ENTIDADES
Y CONCEPTOS"""
porcentaje_coincidencia = 0
for palabra_titulo_querella in palabras_titulo_querella:
if palabra_titulo_querella in palabras_titulo_principal:
porcentaje_coincidencia += 1
porcentaje_coincidencia /= len(palabras_titulo_principal)
if porcentaje_coincidencia >= 0.4:
return True
else:
return False
def build_corpus():
env_path = Path('../') / '.env'
load_dotenv(dotenv_path=env_path)
connection_string = os.getenv("DATABASE_URL")
mongo_client = MongoClient(connection_string)
t4g_database = mongo_client.test
jobs_collection = t4g_database.jobs
jobs = jobs_collection.find()
size = jobs.count()
stemmer = Cistem()
corpus = []
ids = []
for i, job in enumerate(jobs):
if i % 1000 == 0: print(f'{i}/{size}')
indices = []
print(job['_id'])
title = job['title']
_id = job['_id']
ids.append(_id)
text = job['detailed_activities'].strip()
text = ' '.join(text.split())
for index in range(len(text)):
if text[index].isupper() and index > 1 and text[
index - 1] is not " " and text[
index - 2] is not " " and not text.endswith(
text[index]) and text[index + 1] is not " ":
indices.append(index)
for index in reversed(indices):
text = text[:index] + " " + text[index:]
text = re.sub('[^A-Za-zä-üÄ-Ü]', ' ', text)
text = text.lower()
tokenized_text = word_tokenize(text)
words = []
for word in tokenized_text:
stemmed_word = stemmer.stem(word).strip()
if stemmed_word not in stopwords.words(
'german') and word not in stopwords.words(
'german'
) and len(stemmed_word) > 2 and stemmed_word not in [
'it', '3d'
] and stemmed_word not in title:
words.append(stemmed_word)
corpus.append(' '.join(words))
return corpus, ids
## First step: Tokenize each text
from nltk.tokenize import RegexpTokenizer
## Load library for removing stopwords
from nltk.corpus import stopwords
##nltk.download('stopwords') --> First time has to be downloaded
# Import libraries for stemming
from nltk.stem import PorterStemmer
from nltk.tokenize import sent_tokenize
stemmer_ps = PorterStemmer()
from nltk.stem.cistem import Cistem
stemmer_cs = Cistem()
# Import lemmatization libraries
from nltk.stem import WordNetLemmatizer
lemmatizer = WordNetLemmatizer()
#nltk.download('wordnet')
# Load stop words
stop_words = stopwords.words('english')
#print(stop_words[:5])
tokenizer = RegexpTokenizer(r'\w+')
texts_clean = []
texts_aux = []
aux = []
def __init__(self, case_insensitive: BooleanValue()):
self.case_insensitive = case_insensitive
NltkStemmer.__init__(self)
_Cistem.__init__(self, case_insensitive=case_insensitive)
def stem(text):
stemmer = Cistem()
for index, word in enumerate(text):
text[index] = stemmer.stem(word)
return text
import pandas as pd
import nltk
from nltk.stem.cistem import Cistem
# Initialise
tokenizer = nltk.tokenize.TweetTokenizer(strip_handles=True, reduce_len=True)
stemmer = Cistem()
# Read Dataset
tweets = pd.read_csv('data/all_tweets.tsv', sep='\t', header=None)
# Provide rehashed wordlist that is used to filter tweets by topic
keywords = []
index_topic_tweets = []
inp = open("../data/topic_wordlist.txt", "r")
for line in inp.readlines():
line = line.replace('\n', '')
keywords.append(line)
for word in keywords:
word = word.replace('ä', 'ae')
word = word.replace('ö', 'oe')
word = word.replace('ü', 'ue')
word = word.replace('ß', 'ss')
word = word.lower()
keyword = stemmer.stem(word)
for index, row in tweets.iterrows():
# Tokenization
words = tokenizer.tokenize(row[2])
# Remove short tokens
def _get_cis_stemmer(self, case_insensitive):
return Cistem(case_insensitive=case_insensitive)
from nltk.stem.cistem import Cistem
import pandas as pd
import numpy as np
from nltk.tokenize import sent_tokenize
from nltk import pos_tag
from nltk.corpus import stopwords
from sklearn.preprocessing import LabelEncoder
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn import model_selection, naive_bayes, svm
from sklearn.metrics import accuracy_score, f1_score
import re
np.random.seed(500)
stemmer = Cistem()
header_list = ["text_id","text","task_1","task_2"]
regex = re.compile('[,\.!?|#@;:!]')
Corpus = pd.read_csv("train_german.tsv",encoding='latin-1', sep="\t",names=header_list)
Corpus['text'].dropna(inplace=True)
Corpus['text'] = [entry.lower() for entry in Corpus['text']]
Corpus['text'] = [re.sub(r'\w+:\/{2}[\d\w-]+(\.[\d\w-]+)*(?:(?:\/[^\s/]*))*', '', entry) for entry in Corpus['text']]
Corpus['text'] = [entry.replace(".", " ") for entry in Corpus['text']]
Corpus['text'] = [regex.sub(' ', entry) for entry in Corpus['text']]
Corpus['text'] = [entry.split() for entry in Corpus['text']]
#Corpus['text']= [sent_tokenize(entry, language='german') for entry in Corpus['text']]
for index,entry in enumerate(Corpus['text']):
Final_words = []
for word in entry:
if word not in stopwords.words('german') and word.isalpha():
def preprocess(text: str,
bool_to_lowercase=True,
bool_remove_html_tags=True,
bool_remove_links=True,
bool_remove_special_symbols=True,
bool_remove_punctuation=True,
bool_seperate_numbers_from_text=True,
bool_stemming=True,
bool_word_tokenize=True,
bool_sentence_tokenize=False) -> str:
# todo: lemmatization
# transform to lower case
if bool_to_lowercase:
text = text.lower()
# remove symbols, html-tags and links
if bool_remove_html_tags:
logging.debug('Removing html tags')
text = remove_html_tags(text)
if bool_remove_links:
logging.debug('Removing links')
text = remove_links(text)
if bool_remove_special_symbols:
logging.debug('Removing special symbols')
text = remove_special_symbols(text)
# separate numbers from text
if bool_seperate_numbers_from_text:
logging.debug('Separating numbers from text')
text = seperate_numbers_from_text(text)
if bool_remove_punctuation:
logging.debug('Removing punctuation')
text = remove_punctuation_from_text(text)
# tokenize words
if bool_word_tokenize:
logging.debug('Apply word tokenizing')
text_list = word_tokenize(text)
if bool_stemming:
logging.debug('Apply stemming')
text_list = [
Cistem(case_insensitive=False).stem(token)
for token in text_list
]
# remove stopwords
logging.debug('Removing stopwords')
text = remove_stop_words(text_list)
if bool_sentence_tokenize:
logging.debug('Apply sentence tokenizing')
text = sent_tokenize(text, language='german')
return text
if args.log_to_file:
log_file_name = args.target.strip() + '.log'
else:
log_file_name = None
logging.basicConfig(
filename=log_file_name,
format='%(asctime)s : %(levelname)s : %(message)s',
level=logging.INFO)
logging.info("Using up to {} CPUs".format(args.threads))
if args.lemmatize:
logging.info("Perform lemmatisation using nltk")
nlp_de = stanfordnlp.Pipeline(lang="de", processors="tokenize,lemma", use_gpu=True)
if args.stem:
logging.info("Perform stemming using nltk''s CISTEM stemmer")
stemmer = Cistem()
stop_words = stopwords.words('german')
# get corpus sentences
class CorpusSentences(object):
def __init__(self, directory_name: str, chunk_size: int = 10000):
self.directory_name = directory_name
self.n = chunk_size
def __iter__(self):
for file_name in os.listdir(self.directory_name):
logging.info("Use corpus file %s " % file_name)
with codecs.open(os.path.join(self.directory_name, file_name), encoding="utf-8") as f:
while True:
def wordCount(data, dictOutput, catList):
# lade die stopwords
stopwords = load_stopwords()
# Create a new dictionary for the output
outList = collections.OrderedDict()
# Number of non-dictionary words
nonDict = 0
# Convert to lowercase
data = data.lower().replace("\n", " ")
# Tokenize and create a frequency distribution
tokenizer = RegexpTokenizer(r'\w+')
tokens = tokenizer.tokenize(data)
fdist = nltk.FreqDist(tokens)
wc = len(tokens)
# Using the Cistem stemmer for wildcards, create a stemmed version of the data
# Cistem: needed for german words/stemming
cistem = Cistem()
# wenn ein Wort/Token in den stopwords vorkommt, ignoriere dieses
# ansonsten: speichere das gestemmte Wort in der Liste
stems = [cistem.stem(word) for word in tokens if word not in stopwords and len(word) > 0]
fdist_stem = nltk.FreqDist(stems)
# Access categories and populate the output dictionary with keys
for cat in catList:
outList[cat[0]] = 0
# Dictionaries are more useful
fdist_dict = dict(fdist)
fdist_stem_dict = dict(fdist_stem)
# Number of classified words
classified = 0
for key in dictOutput:
if "*" in key and key[:-1] in fdist_stem_dict:
classified = classified + fdist_stem_dict[key[:-1]]
for cat in dictOutput[key]:
if cat.isalpha():
outList[cat] = outList[cat] + fdist_stem_dict[key[:-1]]
elif key in fdist_dict:
classified = classified + fdist_dict[key]
for cat in dictOutput[key]:
try:
outList[cat] = outList[cat] + fdist_dict[key]
except KeyError:
pass
# Calculate the percentage of words classified
if wc > 0:
percClassified = (float(classified) / float(wc)) * 100
else:
percClassified = 0
# Return the categories, the words used, the word count, the number of words classified,
# and the percentage of words classified.
return [outList, tokens, wc, classified, percClassified]
#!usr/bin/env python
#coding:utf8
from nltk.tokenize import TweetTokenizer
from nltk.stem.cistem import Cistem
from nltk.corpus import stopwords
import nltk
from sklearn.feature_extraction.text import CountVectorizer
import matplotlib.pyplot as plt
from wordcloud import WordCloud
from pathlib import Path
from sklearn.cluster import KMeans
from sklearn.manifold import TSNE
nltk.download('stopwords')
tknzr= TweetTokenizer()
stemmer = Cistem(True)
file_in = open("../data/postillon.txt", "r")
file_out = open("../build/preprocessed/postillon_stem.txt", "w")
for line in file_in:
tokenized = tknzr.tokenize(line)
for word in tokenized:
if word in stopwords.words('german'):
tokenized.remove(word)
word = stemmer.stem(word)
token_text = " ".join(tokenized)
file_out.write(token_text+'\n')
file_in.close()
file_out.close()
data = open("../build/preprocessed/postillon_stem.txt", "r")
vectorizer = CountVectorizer(max_features=1000, ngram_range=(1, 3))
