-
Notifications
You must be signed in to change notification settings - Fork 0
/
utils.py
241 lines (214 loc) · 9.59 KB
/
utils.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
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
"""
This module contains functions for preprocessing data.
"""
import os
import json
import pickle
import pandas as pd
import numpy as np
import re
import pycountry
import nltk
import torch
# this is for blocking tensorflow -- it reserves all the gpu memory for some reason
# os.environ["CUDA_VISIBLE_DEVICES"] = ""
import spacy
from nltk.stem import WordNetLemmatizer
from tqdm import tqdm
from gensim.parsing.preprocessing import preprocess_string, strip_multiple_whitespaces, remove_stopwords, strip_punctuation
# download stopwords
# nltk.download('stopwords')
# nltk.download('punkt')
from nltk.corpus import stopwords
from nltk.util import ngrams
from sentence_transformers import SentenceTransformer, util
from sklearn.cluster import AgglomerativeClustering
os.environ["TOKENIZERS_PARALLELISM"] = "true"
# os.environ["CUDA_VISIBLE_DEVICES"] = "1"
class TextProcessor():
"""Class for text processing."""
def __init__(self, filters=None, my_lemmatizer='spacy'):
"""Initialize the class."""
if filters is not None:
self.filters = filters
else:
self.filters = [
lambda x: x.lower(),
self.preprocess,
lambda x: re.sub(r'([^\w\s\\.,]|_)', ' ', x).strip(),
strip_punctuation,
strip_multiple_whitespaces
]
if torch.cuda.is_available():
self.device = f'cuda:{0}'
else:
self.device = 'cpu'
print(f'Using device: {self.device}')
self.cwd = os.getcwd()
self.embedder = SentenceTransformer(f'all-mpnet-base-v2', device=self.device, cache_folder=self.cwd)
self.spacy_model = spacy.load("en_core_web_sm")
self.lemmatizer_to_use = my_lemmatizer
if self.lemmatizer_to_use == 'spacy':
self.lemmatizer = self.spacy_model
else:
self.lemmatizer = WordNetLemmatizer()
self.bioclean = lambda t: re.sub('[.,?;*!%^&_+():-\[\]{}]', '',
t.replace('"', '').replace('/', ' ').replace('\\', '').replace("'",
'').strip().lower())
self.input_embeddings = 'graph_embeddings_with_L6_21_12_2022.p'
self.embeddings = self.load_embeddings()
self.node2idx = {key: idx for idx, key in enumerate(self.embeddings.keys())}
self.idx2node = {v: k for k, v in self.node2idx.items()}
# convert self.input_embeddings to a tensor
self.embeddings = torch.tensor(list(self.embeddings.values()), device=self.device)
my_langs = [
'de', 'it', 'cs', 'da', 'lv', 'es', 'fr', 'bg', 'pl', 'nl', 'el', 'fi', 'sv', 'ro', 'ga', 'hu',
'sk', 'hr', 'pt', 'no', 'sl', 'lt', 'lb', 'et', 'mt', 'so', 'he', 'tr', 'ru', 'th',
'fa', 'ar', 'hi', 'af', 'sq', 'sw', 'ca', 'zh', 'mk', 'ko', 'ur', 'ml', 'vi', 'uk', 'id', 'bn', 'tl', 'ja'
]
# add self.my_stopwords to the list of stopwords
self.my_stopwords = set(stopwords.words('english'))
self.my_stopwords.add('et al')
self.my_stopwords.add('al')
self.my_stopwords.add('et')
self.my_stopwords.add('the')
self.my_stopwords.add('update')
self.my_stopwords.add('recent')
self.my_stopwords.add('method')
self.my_stopwords.add('different')
self.my_stopwords.add('conclusion')
self.my_stopwords.add('review')
self.my_stopwords.add('case')
self.my_stopwords.add('case study')
self.my_stopwords.add('study')
self.my_stopwords.add('an')
self.my_stopwords.add('overview')
self.my_stopwords.add('approach')
self.my_stopwords.add('view')
self.my_stopwords.add('key')
self.my_stopwords.add('analysis')
self.my_stopwords.add('trend')
self.my_stopwords.add('general')
self.my_stopwords.add('classic')
self.my_stopwords.add('model')
self.my_stopwords.add('step')
self.my_stopwords.add('each')
self.my_stopwords.add('amount')
self.my_stopwords.add('>')
self.my_stopwords.add('<')
self.my_stopwords.add('interest')
self.my_stopwords.add("publisher's")
self.my_stopwords.add("ethic")
self.my_stopwords.add("approval")
self.my_stopwords.add("additional")
self.my_stopwords.add("file")
self.my_stopwords.add("supplementary")
self.my_stopwords.add("supplementary material")
self.my_stopwords.add("author")
self.my_stopwords.add("publisher")
self.my_stopwords.add("future research")
self.my_stopwords.add("future work")
self.my_stopwords.add("work")
self.my_stopwords.add("future")
self.my_stopwords.add("impact")
self.my_stopwords.add("literature")
self.my_stopwords.add("goal")
self.my_stopwords.add("scope")
self.my_stopwords.add("definition")
self.my_stopwords.add("cost")
self.my_stopwords.add("challenge")
self.my_stopwords.add("objective")
self.my_stopwords.add("application")
self.my_stopwords.add("scope")
self.my_stopwords.add("present")
self.my_stopwords.add("status")
self.my_stopwords.add("co")
self.my_stopwords.add("-")
self.my_stopwords.add(":")
self.my_stopwords.add("outlook")
self.my_stopwords.add("potential")
self.my_stopwords.add("united states")
self.my_stopwords.add("france")
self.my_stopwords.add("greece")
self.my_stopwords.add("india")
self.my_stopwords.add("germany")
self.my_stopwords.add("project")
self.my_stopwords.add("product")
self.my_stopwords.add("china")
self.my_stopwords.add("japan")
self.my_stopwords.add("south korea")
self.my_stopwords.add("part ii")
self.my_stopwords.add("brazil")
self.my_stopwords.add("new")
self.my_stopwords.add("background")
self.my_stopwords.add("datum")
self.my_stopwords.add("acknowledgement")
self.my_stopwords.add("consent")
self.my_stopwords.add("funding")
self.my_stopwords.add("creation")
self.my_stopwords.add("job")
self.my_stopwords.add("part")
self.my_stopwords.add('comprehensive')
self.my_stopwords.add('survey')
self.my_stopwords.add('research')
self.my_stopwords.add('introduction')
self.my_stopwords.add('discussion')
self.my_stopwords.add('description')
self.my_stopwords.update(my_langs)
self.my_stopwords.update([cntr.alpha_2.lower() for cntr in pycountry.countries])
self.my_stopwords.update([cntr.name.lower() for cntr in pycountry.countries])
self.my_stopwords.update([cntr.alpha_3.lower() for cntr in pycountry.countries])
def load_embeddings(self):
with open(self.input_embeddings, 'rb') as fin:
embeddings = pickle.load(fin)
return embeddings
def preprocess(self, x):
"""Preprocess text."""
return re.sub(r'\s+', ' ', re.sub(r'&.*?;(?:\w*;|#|/?(?:span|p|strong))*', ' ', re.sub(r'<.*?>', ' ', x))).strip()
def wordnet_lemmatize(self, x):
return self.lemmatizer.lemmatize(x)
def spacy_lemmatizer(self, x):
spacy_doc = self.lemmatizer(x)
return ' '.join([token.lemma_ for token in spacy_doc])
def get_spacy_doc(self, x):
return self.spacy_model(x)
def preprocess_text(self, x):
x = ' '.join([tok for tok in preprocess_string(x, filters=self.filters) if tok not in self.my_stopwords])
# lemmatize
if self.lemmatizer_to_use == 'spacy':
x = self.spacy_lemmatizer(x)
else:
x = ' '.join(self.wordnet_lemmatize(tok) for tok in x.split())
return x
def get_ngrams(self, x, k):
return [' '.join(ng) for ng in ngrams(sequence=nltk.word_tokenize(x), n=k)]
def retrieve_similar_nodes(self, query, k):
try:
query_embedding = self.embedder.encode(query, convert_to_tensor=True, device=self.device, show_progress_bar=False)
except RuntimeError:
print('Error in encoding query')
return []
hits = util.semantic_search(query_embedding, self.embeddings, top_k=k, query_chunk_size=1000)
# unpack hits and convert to nodes
return [[(q, self.idx2node[h['corpus_id']]) for h in hit if h['score'] >= 0.8] for q, hit in zip(query, hits)]
def cluster_kws(self, corpus_words, thresh):
# embedder = SentenceTransformer(self.sentence_transformer_name, device=f'cuda:{self.device_id}')
print('Embedding corpus words...')
corpus_embeddings = self.embedder.encode(corpus_words)
# Normalize the embeddings to unit length
corpus_embeddings = corpus_embeddings / np.linalg.norm(corpus_embeddings, axis=1, keepdims=True)
clustering_model = AgglomerativeClustering(
n_clusters=None,
distance_threshold=thresh,
linkage='average',
affinity='cosine'
)
print('Clustering...')
clustering_model.fit(corpus_embeddings)
cluster_assignment = clustering_model.labels_
my_clustered_sentences = {}
for sentence_id, cluster_id in enumerate(cluster_assignment):
if cluster_id not in my_clustered_sentences:
my_clustered_sentences[cluster_id] = []
my_clustered_sentences[cluster_id].append(corpus_words[sentence_id])
return my_clustered_sentences