forked from nadre/clickbait_detection
-
Notifications
You must be signed in to change notification settings - Fork 0
/
clickbait_classifier.py
248 lines (180 loc) · 7.8 KB
/
clickbait_classifier.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
import json
import random
import sklearn.preprocessing
import numpy as np
import scipy.sparse as sparse
import time
from dateutil import parser
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import cross_val_score
from sklearn.utils import shuffle
from sklearn.feature_extraction.text import CountVectorizer
from sklearn.metrics import roc_auc_score, confusion_matrix, classification_report
from sklearn.externals import joblib
def timestamp_to_hour(timestamp):
return parser.parse(timestamp).hour
def timestamp_to_weekday(timestamp):
return parser.parse(timestamp).weekday()
def one_hot_encode(arr):
label_binarizer = sklearn.preprocessing.LabelBinarizer()
label_binarizer.fit(range(max(arr)+1))
return np.array(label_binarizer.transform(arr))
def vectorize_data(data, vocabs={}):
num_instances = len(data)
vectorized_data = sparse.csr_matrix((num_instances, 1))
text_fields = ['targetTitle', 'targetDescription', 'targetKeywords']
for field_name in text_fields:
if field_name in vocabs.keys():
field_vocab = vocabs[field_name]
field_vector, _ = vectorize_text_field(data, field_name, vocab=field_vocab)
else:
field_vector, field_vocab = vectorize_text_field(data, field_name)
vocabs[field_name] = field_vocab
vectorized_data = add_feature(vectorized_data, field_vector)
post_hours = [timestamp_to_hour(x['postTimestamp']) for x in data]
post_hours = one_hot_encode(post_hours)
vectorized_data = add_feature(vectorized_data, post_hours)
post_weekdays = [timestamp_to_weekday(x['postTimestamp']) for x in data]
post_weekdays = one_hot_encode(post_weekdays)
vectorized_data = add_feature(vectorized_data, post_weekdays)
num_paragraphs = np.array([len(x['targetParagraphs']) for x in data]).reshape((num_instances, 1))
vectorized_data = add_feature(vectorized_data, num_paragraphs)
has_media = np.array([len(x['postMedia']) == 0 for x in data]).reshape((num_instances, 1))
vectorized_data = add_feature(vectorized_data, has_media)
paragraph_len = np.array([len(' '.join(x['targetParagraphs']).split(' ')) for x in data]).reshape(
(num_instances, 1))
vectorized_data = add_feature(vectorized_data, paragraph_len)
return vectorized_data, vocabs
def vectorize_text_field(data, field_name, vocab=None):
if vocab is not None:
vectorizer = CountVectorizer(min_df=1, binary=True, ngram_range=(1, 5), vocabulary=vocab)
else:
vectorizer = CountVectorizer(min_df=1, binary=True, ngram_range=(1, 5))
if type(data[0][field_name]) == list:
corpus = [' '.join(x[field_name]) for x in data]
else:
corpus = [x[field_name] for x in data]
vectorized_field = vectorizer.fit_transform(corpus)
return vectorized_field, vectorizer.vocabulary_
def add_feature(feature_set, feature):
return sparse.hstack((feature_set, feature))
def check_data_label_alignment(data, labels):
for i in range(len(data)):
if data[i]['id'] != labels[i][0]:
return False
return True
def balance_data(data, labels):
"""
Assuming there are more negative then positive samples!
"""
num_samples = len(labels)
num_pos_samples = sum([x for x in labels if x])
num_neg_samples = num_samples - num_pos_samples
imbalance = num_neg_samples - num_pos_samples
indexes_of_pos_samples = [i for i, x in enumerate(labels) if x]
for n in range(imbalance):
idx = random.choice(indexes_of_pos_samples)
data.append(data[idx])
labels.append(labels[idx])
return data, labels
def train_and_save_clf(train_data, train_labels, file_path, cross_val=True):
clf = RandomForestClassifier(n_estimators=10)
clf = clf.fit(train_data, train_labels)
if cross_val:
scores = cross_val_score(clf, train_data, train_labels)
print('cross val:')
print(scores)
joblib.dump(clf, file_path)
return clf
def load_clf(file_path):
return joblib.load(file_path)
def evaluate_clf(clf, test_data, test_labels):
predicted_labels = clf.predict(test_data)
print('roc auc:')
print(roc_auc_score(predicted_labels, test_labels))
print('confusion:')
print(confusion_matrix(predicted_labels, test_labels))
print('report:')
target_names = ['clickbait', 'no-clickbait']
print(classification_report(predicted_labels, test_labels, target_names=target_names))
return predicted_labels
def load_data(data_dir):
data = []
for line in open(data_dir + 'instances.jsonl'):
data.append(json.loads(line))
return data
def load_labels(data_dir):
labels = []
for line in open(data_dir + 'truth.jsonl'):
labels.append(json.loads(line))
labels = [(x['id'], x['truthClass'] == 'clickbait') for x in labels]
return labels
def load_and_prepare_data(data_dir):
data = load_data(data_dir)
labels = load_labels(data_dir)
if not check_data_label_alignment(data, labels):
raise Exception
labels = [x[1] for x in labels]
data, labels = balance_data(data, labels)
data, labels = shuffle(data, labels)
return data, labels
# https://stackoverflow.com/questions/27050108/convert-numpy-type-to-python/27050186#27050186
class NumpyEncoder(json.JSONEncoder):
def default(self, obj):
if isinstance(obj, np.integer):
return int(obj)
elif isinstance(obj, np.floating):
return float(obj)
elif isinstance(obj, np.ndarray):
return obj.tolist()
else:
return super(NumpyEncoder, self).default(obj)
def info(method):
def timed(*args, **kw):
ts = time.time()
result = method(*args, **kw)
te = time.time()
print('Function', method.__name__, 'time:', round((te -ts)*1000,1), 'ms')
print()
return result
return timed
@info
def train_and_eval(data_dir, holdout=0.2):
data, labels = load_and_prepare_data(data_dir)
vectorized_data, vocabs = vectorize_data(data)
json.dump(vocabs, open(data_dir+'vocabs.json', 'w'), cls=NumpyEncoder)
train_test_split = int(len(labels)*holdout)
vectorized_data = vectorized_data.tocsr()
train_data = vectorized_data[:train_test_split, :]
test_data = vectorized_data[train_test_split:, :]
train_labels = labels[:train_test_split]
test_labels = labels[train_test_split:]
clf = train_and_save_clf(train_data, train_labels, data_dir+'RandomForestClassifier.pickle')
predicted_labels = evaluate_clf(clf, test_data, test_labels)
for i, predicted_clickbait in enumerate(predicted_labels):
if predicted_clickbait and not test_labels[i]:
print(data[train_test_split+i]['targetTitle'])
@info
def load_and_eval(data_dir):
data, labels = load_and_prepare_data(data_dir)
vocabs = json.load(open(data_dir+'vocabs.json', 'r'))
vectorized_data, _ = vectorize_data(data, vocabs)
clf = load_clf(data_dir+'RandomForestClassifier.pickle')
predicted_labels = evaluate_clf(clf, vectorized_data, labels)
# for i, predicted_clickbait in enumerate(predicted_labels):
# if predicted_clickbait and not labels[i]:
# print(data[i]['targetTitle'])
@info
def train(data_dir):
data, labels = load_and_prepare_data(data_dir)
vectorized_data, vocabs = vectorize_data(data)
json.dump(vocabs, open(data_dir+'vocabs.json', 'w'), cls=NumpyEncoder)
np.save(data_dir+'vectorized_data.pickle', vectorized_data)
vectorized_data = vectorized_data.tocsr()
_ = train_and_save_clf(vectorized_data, labels, data_dir+'RandomForestClassifier.pickle')
if __name__ == '__main__':
# data_dir = '/home/neffle/data/clickbait/clickbait17-train-170331/'
data_dir = '/home/xuri3814/data/clickbait17-validation-170616/'
# train_and_eval(data_dir)
train(data_dir)
load_and_eval(data_dir)