CODEWITHSUNDEEP
pythonnlpnltksentiment-analysisvader
Adrian Dolinay
Adrian Dolinay

Reputation: 151

NLTK Vader SentimentIntensityAnalyzer Bigram

For the VADER SentimentIntensityAnalyzer within Python, is there a way to add a bigram rule? I tried updating the lexicon with a two word input, but it did not change the polarity score. Thanks in advance!

from vaderSentiment.vaderSentiment import SentimentIntensityAnalyzer

analyser = SentimentIntensityAnalyzer()

#returns a compound score of -0.296
print(analyser.polarity_scores('no issues'))

analyser.lexicon['no issues'] = 0.0
#still returns a compound score of -0.296
print(analyser.polarity_scores('no issues'))

Upvotes: 1

Views: 851

Answers (1)

Vipul Joshi
Vipul Joshi

Reputation: 48

There is no straightforward way to add bigram to the vader lexicon. This is because vader considers individual tokens for sentiment analysis. However, one can do this using following steps:

  1. Create bigrams as tokens. For example, you can convert the bigram ("no issues") into a token ("noissues").
  2. Maintain a dictionary of polarity of the newly created tokens. {"noissues" : 2}
  3. Then perform additional text processing before passing the text for sentiment score calculation.

Following code accomplishes the above:

allowed_bigrams = {'noissues' : 2} #add more as per your requirement
    
def process_text(text):
    tokens = text.lower().split() # list of tokens
    bigrams = list(nltk.bigrams(tokens)) # create bigrams as tuples of tokens
    bigrams = list(map(''.join, bigrams)) # join each word without space to create new bigram
    bigrams.append('...') # make length of tokens and bigrams list equal
     
    #begin recreating the text
    final = ''
    for i, token in enumerate(tokens):
        b = bigrams[i]
        
        if b in allowed_bigrams:
          join_word = b # replace the word in text by bigram
          tokens[i+1] = '' #skip the next word
        else:
            join_word = token
        final += join_word + ' '
    return final
text  = 'Hello, I have no issues with you'
print (text)
print (analyser.polarity_scores(text))
final = process_text(text)
print (final)
print(analyser.polarity_scores(final))

The output :

Hello, I have no issues with you
{'neg': 0.268, 'neu': 0.732, 'pos': 0.0, 'compound': -0.296}
hello, i have noissues  with you 
{'neg': 0.0, 'neu': 0.625, 'pos': 0.375, 'compound': 0.4588}

Notice in the output, how two words "no" and "issues" have been added together to form bigram "noissues".

Upvotes: 3

Related Questions