from nltk.corpus import wordnet as wordnet
from nltk.corpus import sentiwordnet as swn
import nltk
def wordnet_pos_code(tag):
'''Translation from nltk tags to Wordnet code'''
if tag.startswith('NN'):
return wordnet.NOUN
elif tag.startswith('VB'):
return wordnet.VERB
elif tag.startswith('JJ'):
return wordnet.ADJ
elif tag.startswith('RB'):
return wordnet.ADV
else:
return ''
def pos_tag(sentence):
'''POS tagging of a sentence.'''
tagged_words = []
tokens = nltk.word_tokenize(sentence)
tag_tuples = nltk.pos_tag(tokens)
for (string, tag) in tag_tuples:
token = {'word':string, 'pos':tag}
tagged_words.append(token)
return tagged_words
def word_sense_cdf(word, context, wn_pos):
'''Word sense disambiguation in terms of matching words frequency
between the context each sense's definition. Adapted from
www.slideshare.net/faigg/tutotial-of-sentiment-analysis'''
senses = wordnet.synsets(word, wn_pos)
if len(senses) > 0:
cfd = nltk.ConditionalFreqDist((sense, def_word)
for sense in senses
for def_word in sense.definition().split()
if def_word in context)
best_sense = senses[0]
for sense in senses:
try:
if cfd[sense].max() > cfd[best_sense].max():
best_sense = sense
except:
pass
return best_sense
else:
return None
def word_sense_similarity(word, context, dummy = None):
'''Another word sense disambiguation technique. It's VERY SLOW.
Adapted from: pythonhosted.org/sentiment_classifier'''
wordsynsets = wordnet.synsets(word)
bestScore = 0.0
result = None
for synset in wordsynsets:
for w in nltk.word_tokenize(context):
score = 0.0
for wsynset in wordnet.synsets(w):
sim = wordnet.path_similarity(wsynset, synset)
if(sim == None):
continue
else:
score += sim
if (score > bestScore):
bestScore = score
result = synset
return result
def sentiwordnet_classify(text):
'''Breaks a multi sentence text to separate sentences.
This improves context for the word sense disambiguation.
Returns a class'''
score_tot = 0
score_tot_thr = 0
class_tot = 0
class_tot_thr = 0
sentences = nltk.sent_tokenize(text)
for sentence in sentences:
(score, score_thr) = sentence_score(sentence)
score_tot += score
score_tot_thr += score_thr
#Trust the thresholded value more when classifying
if score_tot_thr != 0:
clss = 'pos' if score_tot_thr > 0 else 'neg'
elif score_tot != 0:
clss = 'pos' if score_tot > 0 else 'neg'
else:
clss = None
return clss
def sentence_score(text, threshold = 0.75, wsd = word_sense_cdf):
'''Classifies a phrase according to sentiment analysis based
on WordNet and SentiWordNet. It also computes a thresholded
score by ignoring strongly objective words.'''
tagged_words = pos_tag(text)
obj_score = 0 # object score
pos_score=0 # positive score
neg_score=0 #negative score
pos_score_thr=0
neg_score_thr=0
for word in tagged_words:
# print word
if 'punct' not in word :
sense = wsd(word['word'], text, wordnet_pos_code(word['pos']))
if sense is not None:
sent = swn.senti_synset(sense.name())
if sent is not None and sent.obj_score() <> 1:
obj_score = obj_score + float(sent.obj_score())
pos_score = pos_score + float(sent.pos_score())
neg_score = neg_score + float(sent.neg_score())
if sent.obj_score() < threshold:
pos_score_thr = pos_score_thr + float(sent.pos_score())
neg_score_thr = neg_score_thr + float(sent.neg_score())
return (pos_score - neg_score, pos_score_thr - neg_score_thr)
