-
Notifications
You must be signed in to change notification settings - Fork 0
/
svm-sem-eval.py
executable file
·144 lines (100 loc) · 5.17 KB
/
svm-sem-eval.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
# /usr/lib/python
import time
import re
from sets import Set
import numpy as np
import pandas as pd
import sklearn.cross_validation
import sklearn.feature_extraction.text
import sklearn.metrics
import sklearn.naive_bayes
from nltk.tokenize import PunktWordTokenizer
from nltk.stem import PorterStemmer, WordNetLemmatizer
from nltk.stem import PorterStemmer, WordNetLemmatizer
from nltk.tokenize import PunktWordTokenizer, WordPunctTokenizer
start_time = time.time()
lemmatizer = WordNetLemmatizer()
stemmatizer = PorterStemmer()
tokenizer = WordPunctTokenizer()
def my_token(s):
my_tokenizer = PunktWordTokenizer()
return my_tokenizer.tokenize(s)
stop_words = Set()
with open('english.stop', 'rb') as stop_w:
for word in stop_w:
stop_words.add(word.decode('utf-8').rstrip())
def sanitizer(tweet):
tweet_trimed = re.sub('http[s]?://(?:[a-zA-Z]|[0-9]|[$-_@.&+]|[!*\(\),]|(?:%[0-9a-fA-F][0-9a-fA-F]))+', 'url',
tweet)
tweet_trimed_hash_at = re.sub(':\)|:-\)|:D|:-\]', 'good', tweet_trimed)
tweet_trimed_hash_at = re.sub(':\(|:-\(|:\[|:\|', 'bad', tweet_trimed_hash_at)
# tweet_trimed_hash_at = re.sub('\d+.\d+|\d+|\d+th', 'number', tweet_trimed_hash_at)
tweet_trimed_hash_at = re.sub('/^\d*\.?\d*$/', 'number', tweet_trimed_hash_at)
tweet_trimed_hash_at = re.sub('/^\d*\,?\d*$/', 'number', tweet_trimed_hash_at)
tweet_trimed_hash_at = re.sub('/^\d*\-?\d*$/', 'number', tweet_trimed_hash_at)
# tweet_trimed_hash_at = re.sub("[\(\)&,.:!?`~-]+", '', tweet_trimed_hash_at)
# tweet_trimed_hash_at = re.sub("(?<=^|(?<=[^a-zA-Z0-9-_\.]))@([A-Za-z]+[A-Za-z0-9]+)", '', tweet_trimed_hash_at)
tweet_trimed_hash_at = re.sub('@[\s]+', 'atuser', tweet_trimed_hash_at)
tweet_token = tokenizer.tokenize(tweet_trimed_hash_at.decode('utf-8'))
tweet_trimed_hash_at = tweet_trimed_hash_at.translate(None, string.punctuation)
tweet_token = tokenizer.tokenize(tweet_trimed_hash_at.decode('ISO-8859-1'))
tweet_sanitized = [word.lower() for word in tweet_token]
tweet_lemmatized = [lemmatizer.lemmatize(word) for word in tweet_sanitized]
# print tweet_lemmatized
# tweet_stemmatized = [stemmatizer.stem(word) for word in tweet_sanitized if word not in stop_words]
# tweet_replaced = [replacer.replace(word) for word in tweet_stemmatized]
# print tweet_stemmatized
# no_punctuation = lowers.translate(None, string.punctuation)
tweet_obradjen = " ".join(tweet_lemmatized)
return tweet_obradjen
categories = ['positive', 'negative']
names = ['label', 'broj', 'datum', 'topic', 'user', 'text']
data_train = pd.read_table('/home/dudulez/sentiment_analysis/trainingandtestdata/training.1600000_posneg.csv', sep=',',
names=names)
data_test = pd.read_table('/home/dudulez/sentiment_analysis/trainingandtestdata/testdata.csv', sep=',', names=names)
# split data into traininig and testing sets. Default is 75% train
# )
# after the split, we have two arrays of arrays
# train, test = data_train,data_test
train = np.array(
data_train.ix[:, names]) # , train1 = sklearn.cross_validation.train_test_split(data_train, train_size= .99999999)
# test, test1 = sklearn.cross_validation.train_test_split(data_test, train_size= .99999999)
test = np.array(data_test.ix[:, names])
train_data, test_data = pd.DataFrame(train, columns=names), pd.DataFrame(test, columns=names)
# vidi da li ovde moze da se promeni nesto
# vectorization is the process of converting all names into a binary vector
from sklearn.feature_extraction.text import TfidfVectorizer
vectorizer = TfidfVectorizer(stop_words=stop_words,
ngram_range=(1, 1),
lowercase=True,
tokenizer=my_token,
preprocessor=sanitizer
)
# vectorizer = sklearn.feature_extraction.CountVectorizer()#encoding=u'utf-8',lowercase=True,
# tokenizer = None, ngram_range(2,3), stop_words = 'english', min_df = 0, max_df = 0.1)
train_matrix = vectorizer.fit_transform(train_data['text'])
test_matrix = vectorizer.transform(test_data['text'])
positive_cases_train = (train_data['label'] == 'positive')
positive_cases_test = (test_data['label'] == 'positive')
negative_cases_train = (train_data['label'] == 'negative')
negative_cases_test = (test_data['label'] == 'negative')
# ratio imbalance
# train
from sklearn.linear_model import SGDClassifier
# classifier = LinearSVC()un
# classifier = sklearn.naive_bayes.MultinomialNB()
# classifier = LogisticRegression(penalty='l2', C=1.0)
classifier = SGDClassifier(alpha=0.00001, l1_ratio=0.015)
classifier.fit(train_matrix, positive_cases_train)
predict_sentiment = classifier.predict(test_matrix)
# predict_probs = classifier.predict_proba(test_matrix)
accuracy = classifier.score(test_matrix, positive_cases_test)
precision, recall, f1, _ = sklearn.metrics.precision_recall_fscore_support(
positive_cases_test, predict_sentiment)
print(" LogisticRegression, no preprocesing, unigram only")
print ("accuracy = ", accuracy)
print (" precision =", precision)
print ("recal = ", recall)
print ("f1 score = ", f1)
end_time = time.time()
print 'Iterations took %f seconds.' % (end_time - start_time)