-
Notifications
You must be signed in to change notification settings - Fork 1
/
cluster_abstracts.py
253 lines (212 loc) · 8.12 KB
/
cluster_abstracts.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
242
243
244
245
246
247
248
249
250
251
252
253
import nltk
import re
import pickle
import argparse
import pandas as pd
from Bio import Entrez
from sklearn.cluster import KMeans
from nltk.stem.snowball import SnowballStemmer
from sklearn.decomposition import TruncatedSVD
from sklearn.pipeline import make_pipeline
from sklearn.preprocessing import Normalizer
from sklearn.feature_extraction.text import TfidfVectorizer
class PubMedClustering:
"""PubMed Abstract Clusterer.
Attributes:
filename: Input filename containing PMIDs.
n_components: Dimensionality of LSA output.
num_clusters: Number of clusters to form.
"""
def __init__(
self,
filename,
n_components=100,
num_clusters=6,
num_words=5,
f_store=False):
""" Sets up PubMed Clusterer. """
self.filename = filename
self.n_components = n_components
self.n_clusters = num_clusters
self.n_words = num_words
self.f_store = f_store
# Getting the NLTK English language stopword list.
self.stopwords = nltk.corpus.stopwords.words('english')
# Stemming the stopwords for consistency.
stemmer = SnowballStemmer("english")
self.stemmed_stopwords = [
stemmer.stem(w) for w in self.stopwords] + [
"'d",
'could',
'might',
'must',
"n't",
'need',
'r',
'sha',
'v',
'wo',
'would']
def retrieve_abstracts(self):
""" Retrieves PubMed abstracts from the NCBI Entrez database. """
with open('data/' + self.filename, 'r') as f:
lines = f.readlines()
self.pmids = [line.rsplit()[0] for line in lines]
Entrez.email = 'spawar3@uncc.edu'
# Fetching PubMed records from the NCBI Entrez DB.
print('Retrieving PubMed abstracts...')
handle = Entrez.efetch(
db="pubmed",
id=','.join(self.pmids),
rettype="xml",
retmode="text")
records = Entrez.read(handle)
abstracts = []
# Extracting abstract text from PubMed articles.
for article in records['PubmedArticle']:
if 'Abstract' in article['MedlineCitation']['Article'].keys():
abstract = article['MedlineCitation']['Article']['Abstract']
abstract_text = abstract['AbstractText'][0]
abstracts.append(abstract_text)
else:
# Using article title as a proxy for abstract when abstract is
# missing.
abstracts.append(
article['MedlineCitation']['Article']['ArticleTitle'])
return abstracts
def tokenize(self, text, stem=True):
""" Preprocesses the text data. """
# Tokenizing the abstracts first by sentence, then by word.
tokens = [word for sent in nltk.sent_tokenize(
text) for word in nltk.word_tokenize(sent)]
filtered_tokens = []
# Filtering out any tokens not containing letters (e.g. numeric tokens,
# raw punctuation, etc.)
for token in tokens:
if re.search('[a-zA-Z]', token):
filtered_tokens.append(token)
if stem:
# Stemming the tokens.
stemmer = SnowballStemmer("english")
tokens = [stemmer.stem(t) for t in filtered_tokens]
return tokens
else:
return filtered_tokens
def cluster_abstracts(self):
""" Clusters abstracts using K-Means. """
self.abstracts = self.retrieve_abstracts()
# Converting abstracts to a matrix of TF-IDF features.
print('Extracting tf-idf features...')
tfidf_vectorizer = TfidfVectorizer(
stop_words=self.stemmed_stopwords,
use_idf=True,
tokenizer=self.tokenize,
ngram_range=(1, 3))
tfidf_matrix = tfidf_vectorizer.fit_transform(self.abstracts)
self.terms = tfidf_vectorizer.get_feature_names()
# Performing dimensionality reduction using LSA and normalizing the
# results.
print('Performing dimensionality reduction...')
svd = TruncatedSVD(self.n_components)
normalizer = Normalizer(copy=False)
lsa = make_pipeline(svd, normalizer)
self.X = lsa.fit_transform(tfidf_matrix)
# Applying K-Means to obtain clusters
print('Clustering abstracts...')
self.km = KMeans(n_clusters=self.n_clusters, n_jobs=-1, n_init=10)
self.km.fit(self.X)
if self.f_store:
print('Storing model to disk...')
with open('models/clusterer_model.pkl', 'wb') as f:
pickle.dump(self.km, f)
self.clusters = self.km.labels_.tolist()
self.original_space_centroids = svd.inverse_transform(
self.km.cluster_centers_)
def get_top_terms(self):
""" Finds top terms or "composition" of clusters. """
totalvocab_stemmed = []
totalvocab_tokenized = []
for i in self.abstracts:
allwords_stemmed = self.tokenize(i, stem=True)
totalvocab_stemmed.extend(allwords_stemmed)
allwords_tokenized = self.tokenize(i, stem=False)
totalvocab_tokenized.extend(allwords_tokenized)
# Creating a mapping from stems to their original forms/tokens
self.stem_to_token = {}
for i, term in enumerate(totalvocab_stemmed):
if term not in self.stem_to_token.keys():
self.stem_to_token[term] = totalvocab_tokenized[i]
# Sorting to identify which are the top words that are nearest to the
# cluster centroid (to extract cluster "composition")
self.order_centroids = self.original_space_centroids.argsort()[:, ::-1]
self.cluster_top_terms = {}
# Getting top terms per cluster
for i in range(self.n_clusters):
cluster_terms = []
for ind in self.order_centroids[i, :self.n_words]:
term = ''
for token in self.terms[ind].split():
term += self.stem_to_token[token] + ' '
cluster_terms.append(term)
self.cluster_top_terms[str(i)] = cluster_terms
def store_output(self):
""" Stores the clustering output to a text file. """
self.cluster_compositions = [', '.join(
self.cluster_top_terms[str(cluster)]) for cluster in self.clusters]
self.pubmed_articles = {
'pmid': self.pmids,
'abstract': self.abstracts,
'cluster': self.clusters,
'cluster_composition': self.cluster_compositions}
self.pm_article_df = pd.DataFrame(self.pubmed_articles)
print('Storing clustering output...')
self.pm_article_df.to_csv(
'output/output_clusters.txt',
columns=[
'pmid',
'cluster_composition'],
header=False,
index=False,
sep=' ')
def parse_arguments():
""" Parses command-line arguments. """
parser = argparse.ArgumentParser()
parser.add_argument(
'filename',
type=str,
help='name of the file containing PubMed article IDs')
parser.add_argument(
'--lsa',
dest='n_components',
type=int,
default=100,
help='dimensionality of latent semantic analysis output')
parser.add_argument(
'--num_clusters',
dest='num_clusters',
type=int,
default=6,
help='number of clusters to form')
parser.add_argument(
'--num_words',
dest='num_words',
type=int,
default=5,
help='number of top words per cluster')
parser.add_argument(
'--store_model',
action='store_true',
help='persist the model on disk')
args = parser.parse_args()
return args
if __name__ == "__main__":
args = parse_arguments()
clusterer = PubMedClustering(
args.filename,
args.n_components,
args.num_clusters,
args.num_words,
args.store_model)
clusterer.cluster_abstracts()
clusterer.get_top_terms()
clusterer.store_output()