Sentiment Analysis very slow in mac

[janeshdev]

Hi, I am using flair for sentiment analysis using en-sentiment text classifier and I am running for 2000 sentences and it's very slow. The code is running for more than 3 hours now and still not finished. The code I am using is as follows:

from flair.models import TextClassifier
from flair.data import Sentence
classifier = TextClassifier.load('en-sentiment')

# Create a function 
def sentiment_analysis(text):
    tt = Sentence(text)
    classifier.predict(tt)
    if tt.labels[0].value == 'POSITIVE':
        multiplier = 1
    else: 
        multiplier = -1  
    sentiment = multiplier * tt.labels[0].score 
    return(';;'.join([tt.labels[0].value, repr(sentiment)]))

df['sentiment'] = df.customer_comments.apply(sentiment_analysis).str.split(";;")

Any tips or ideas to speed up this implementation?

Also can you please let me know how is the sentiment calculated in en-sentiment ? I am using this to find sentiment for customer reviews. Would it make sense to use on customer reviews?

Thanks for the wonderful package.

[alanakbik]

Hello @janeshdev how long are your documents and do you have a GPU?

[alanakbik]

The current sentiment model is trained using English movie reviews, so it will work best for such text. We are in the process of training better sentiment models over a mix of domains, which should be added very soon. We can add a 'normal' and a 'fast' variant for CPU-only setups.

[janeshdev]

Hello @janeshdev how long are your documents and do you have a GPU?

Hi I don't have a GPU and there are approximately 35000 comments that I was trying to use Sentiment analysis on. Do you have a list of sentiment models that are on pipeline? I would love to see sentiment analysis based on customer comments on service / product etc.

[alanakbik]

Yes, I am currently training models over the datasets as described in #1545 which includes a lot of product reviews, tweets and other data. I will push them very soon and let you know!

[alanakbik]

@janeshdev I am currently training sentiment models (see #1545) that includes Amazon reviews across many categories. They will be made available soon!

[janeshdev]

Hi @alanakbik are the new sentiment models ready to use?

[alanakbik]

Just merged #1613. On master branch, two new (much better) models are now available.