-
Notifications
You must be signed in to change notification settings - Fork 31
/
extract_baseline_features.py
247 lines (229 loc) · 10.3 KB
/
extract_baseline_features.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
from nltk.tokenize import TweetTokenizer, RegexpTokenizer
from nltk.stem.wordnet import WordNetLemmatizer
from nltk import ngrams, pos_tag
from collections import Counter
import numpy as np
import vocab_helpers as helper
def count_apparitions(tokens, list_to_count_from):
total_count = 0.0
for affirmative in list_to_count_from:
total_count += tokens.count(affirmative)
return total_count
def get_features1(tweets, subj_dict):
print("Getting features type 1...")
features = []
tknzr = TweetTokenizer(preserve_case=False, reduce_len=True, strip_handles=True)
lemmatizer = WordNetLemmatizer()
# Take positive and negative noun/verb phrases as features
for tweet in tweets:
feature_list = [0.0] * 6
tokens = tknzr.tokenize(tweet)
pos = pos_tag(tokens)
pos = [p for p in pos if 'VB' in p[1] or 'NN' in p[1]]
for p in pos:
stemmed = lemmatizer.lemmatize(p[0], 'v')
stemmed = lemmatizer.lemmatize(stemmed)
if 'VB' in p[1] and stemmed in subj_dict:
if 'verb' in subj_dict[stemmed]:
if 'positive' in subj_dict[stemmed]['verb']:
feature_list[0] += 1.0
if 'negative' in subj_dict[stemmed]['verb']:
feature_list[1] += 1.0
elif 'anypos' in subj_dict[stemmed]:
if 'positive' in subj_dict[stemmed]['anypos']:
feature_list[0] += 1.0
if 'negative' in subj_dict[stemmed]['anypos']:
feature_list[1] += 1.0
if 'NN' in p[1] and stemmed in subj_dict:
if 'noun' in subj_dict[stemmed]:
if 'positive' in subj_dict[stemmed]['noun']:
feature_list[2] += 1.0
if 'negative' in subj_dict[stemmed]['noun']:
feature_list[3] += 1.0
elif 'anypos' in subj_dict[stemmed]:
if 'positive' in subj_dict[stemmed]['anypos']:
feature_list[2] += 1.0
if 'negative' in subj_dict[stemmed]['anypos']:
feature_list[3] += 1.0
# Derive features from punctuation
feature_list[4] += count_apparitions(tokens, helper.punctuation)
# Take the number of strong negations as a feature
feature_list[5] += count_apparitions(tokens, helper.strong_negations)
features.append(feature_list)
print("Done.")
return features
def get_features2(tweets, subj_dict):
print("Getting features type 2...")
features = []
tknzr = TweetTokenizer(preserve_case=True, reduce_len=False, strip_handles=False)
lemmatizer = WordNetLemmatizer()
for tweet in tweets:
feature_list = [0.0] * 5
tokens = tknzr.tokenize(tweet)
# Take the number of positive and negative words as features
for word in tokens:
stemmed = lemmatizer.lemmatize(word, 'v')
stemmed = lemmatizer.lemmatize(stemmed)
if stemmed in subj_dict:
dictlist = []
for word in subj_dict[stemmed]:
dictlist.extend(subj_dict[stemmed][word])
if 'strongsubj' in dictlist:
value = 1.0
else:
value = 0.5
if 'positive' in dictlist:
feature_list[0] += value
elif 'negative' in dictlist:
feature_list[1] += value
# Take the report of positives to negatives as a feature
if feature_list[0] != 0.0 and feature_list[1] != 0.0:
feature_list[2] = feature_list[0] / feature_list[1]
# Derive features from punctuation
feature_list[2] += count_apparitions(tokens, helper.punctuation)
# Take strong negations as a feature
feature_list[3] += count_apparitions(tokens, helper.strong_negations)
# Take strong affirmatives as a feature
feature_list[4] += count_apparitions(tokens, helper.strong_affirmatives)
features.append(feature_list)
print("Done.")
return features
def get_features3(tweets, subj_dict):
print("Getting features type 3...")
features = []
tknzr = TweetTokenizer(preserve_case=False, reduce_len=True, strip_handles=False)
lemmatizer = WordNetLemmatizer()
# Take positive and negative noun/verb phrases as features
for tweet in tweets:
feature_list = [0.0] * 8
tokens = tknzr.tokenize(tweet)
pos = pos_tag(tokens)
pos = [p for p in pos if 'VB' in p[1] or 'NN' in p[1]]
for p in pos:
stemmed = lemmatizer.lemmatize(p[0], 'v')
stemmed = lemmatizer.lemmatize(stemmed)
if 'VB' in p[1] and stemmed in subj_dict:
if 'verb' in subj_dict[stemmed]:
if 'strongsubj' in subj_dict[stemmed]['verb']:
value = 1.0
else:
value = 0.5
if 'positive' in subj_dict[stemmed]['verb']:
feature_list[0] += value
elif 'negative' in subj_dict[stemmed]['verb']:
feature_list[1] += value
elif 'anypos' in subj_dict[stemmed]:
if 'strongsubj' in subj_dict[stemmed]['anypos']:
value = 1.0
else:
value = 0.5
if 'positive' in subj_dict[stemmed]['anypos']:
feature_list[0] += value
elif 'negative' in subj_dict[stemmed]['anypos']:
feature_list[1] += value
if 'NN' in p[1] and stemmed in subj_dict:
if 'noun' in subj_dict[stemmed]:
if 'strongsubj' in subj_dict[stemmed]['noun']:
value = 1.0
else:
value = 0.5
if 'positive' in subj_dict[stemmed]['noun']:
feature_list[2] += value
elif 'negative' in subj_dict[stemmed]['noun']:
feature_list[3] += value
elif 'anypos' in subj_dict[stemmed]:
if 'strongsubj' in subj_dict[stemmed]['anypos']:
value = 1.0
else:
value = 0.5
if 'positive' in subj_dict[stemmed]['anypos']:
feature_list[2] += value
elif 'negative' in subj_dict[stemmed]['anypos']:
feature_list[3] += value
# Take the report of positives to negatives as a feature
if (feature_list[0] + feature_list[2]) != 0.0 and (feature_list[1] + feature_list[3]) != 0.0:
feature_list[4] = (feature_list[0] + feature_list[2]) / (feature_list[1] + feature_list[3])
# Derive features from punctuation
feature_list[5] += count_apparitions(tokens, helper.punctuation)
# Take strong negations as a feature
feature_list[6] += count_apparitions(tokens, helper.strong_negations)
# Take strong affirmatives as a feature
feature_list[7] += count_apparitions(tokens, helper.strong_affirmatives)
features.append(feature_list)
print("Done.")
return features
def get_ngram_list(tknzr, text, n):
tokens = tknzr.tokenize(text)
tokens = [t for t in tokens if not t.startswith('#')]
tokens = [t for t in tokens if not t.startswith('@')]
ngram_list = [gram for gram in ngrams(tokens, n)]
return ngram_list
def get_ngrams(tweets, n):
unigrams = Counter()
bigrams = Counter()
trigrams = Counter()
regexp_tknzr = RegexpTokenizer(r'\w+')
tweet_tknzr = TweetTokenizer()
for tweet in tweets:
tweet = tweet.lower()
# Get the unigram list for this tweet and update the unigram counter
unigram_list = get_ngram_list(tweet_tknzr, tweet, 1)
unigrams.update(unigram_list)
# Get the bigram list for this tweet and update the bigram counter
if n > 1:
bigram_list = get_ngram_list(regexp_tknzr, tweet, 2)
bigrams.update(bigram_list)
# Get the trigram list for this tweet and update the trigram counter
if n > 2:
trigram_list = get_ngram_list(regexp_tknzr, tweet, 3)
trigrams.update(trigram_list)
# Update the counters such that each n-gram appears at least min_occurence times
min_occurence = 2
unigram_tokens = [k for k, c in unigrams.items() if c >= min_occurence]
# In case using just unigrams, make the bigrams and trigrams empty
bigram_tokens = trigram_tokens = []
if n > 1:
bigram_tokens = [k for k, c in bigrams.items() if c >= min_occurence]
if n > 2:
trigram_tokens = [k for k, c in trigrams.items() if c >= min_occurence]
return unigram_tokens, bigram_tokens, trigram_tokens
def create_ngram_mapping(unigrams, bigrams, trigrams):
ngram_map = dict()
all_ngrams = unigrams
all_ngrams.extend(bigrams)
all_ngrams.extend(trigrams)
for i in range(0, len(all_ngrams)):
ngram_map[all_ngrams[i]] = i
return ngram_map
def get_ngram_features_from_map(tweets, ngram_map, n):
regexp_tknzr = RegexpTokenizer(r'\w+')
tweet_tknzr = TweetTokenizer()
features = []
for tweet in tweets:
feature_list = [0] * np.zeros(len(ngram_map))
tweet = tweet.lower()
ngram_list = get_ngram_list(tweet_tknzr, tweet, 1)
if n > 1:
ngram_list += get_ngram_list(regexp_tknzr, tweet, 2)
if n > 2:
ngram_list += get_ngram_list(regexp_tknzr, tweet, 3)
for gram in ngram_list:
if gram in ngram_map:
feature_list[ngram_map[gram]] += 1.0
features.append(feature_list)
return features
def get_ngram_features(tweets, n):
print("Getting n-gram features...")
unigrams = []
bigrams = []
trigrams = []
if n == 1:
unigrams, _, _ = get_ngrams(tweets, n)
if n == 2:
unigrams, bigrams, _ = get_ngrams(tweets, n)
if n == 3:
unigrams, bigrams, trigrams = get_ngrams(tweets, n)
ngram_map = create_ngram_mapping(unigrams, bigrams, trigrams)
features = get_ngram_features_from_map(tweets, ngram_map, n)
print("Done.")
return ngram_map, features