NLP and sentiment analysis without labels?

@jeremy @jcatanza - have customer opened case descriptions only. No lables availables against it .I want to predict the sentiment like opinion,tone of the customer,suggestion, a complaint, nonsense/spam, a threat.
How can we deal this and which library is best to deal without a label in the in customer service area.
Please help me with Good library names and piece of code

first, don’t tag jeremy unless there is an incredibly good reason that only jeremy can answer the question.

second, you need to train on labeled data, so you can either find some other labeled data which looks similar to yours and train on that, or label enough of your own data.

1 Like

@joedockrill - I apologize for it.Surely will not do that.
Without labels Can’t we get sentiment scores the way new libraries like Textblop,Vader giving polarity scores with subject.

Anything which will work out of the box like that has to be pre trained, but is probably trained in a more generic way than you need, for example textblob tells you if it’s objective Vs subjective which is useless information for customer services as everything you’re dealing with is subjective.

Do your own research into some of these libraries but ultimately I suspect the answer which will provide the best results is “label some of your data”.

1 Like

@joedockrill - Thank you Very much for detailed explanation!

While I agree with @joedockrill that getting some labelled examples is probably the best way to go if your texts are very specific to your domain, there are some pre-trained models that you can use even without fine-tuning on your data.

Huggingface just released a new zero shot classification pipeline that also works for sentiment classification: see this Colab notebook

And on their model hub you can find plenty of pre-trained sentiment classification models:


Agree with both @joedockrill and @stefan-ai and I’ll add that I have read there are some interesting tools now for labelling your own data more efficiently, Snorkel is one that was released recently

@stefan-ai Thanks for prompt reply on my query.
:+1: :clap: :pray: