Sentiment anlaysis using NLP

nlp

#1

Hi Folks,

I am working on a NLP project (using python) in which we are predicting the sentiment of people conversation that happen on chat pod. Basically we are classifying sentiment to be either positive and negative which could help us to judge the depression of a person. We have successfully build this model and algorithm is able to predict the correct sentiments.

Saying so, we have further requirements on this, now customer is looking for how to judge whether person is feeling anxious, stressed, angry, depressed. I could think of building different kind of corpus namely anxious, stressed etc based on which classifier will classify the sentiment mode.

Again that’ my take on it, I am looking for a better solution or logic which can be implemented. Please feel free to comment.

Thanks,
Shubham


#2

I would like to tell what I know…
The “further requirements” can also be called as Emotion Classification. It can be implemented using simply categorizing & counting the words which fall into each of these categories. We can have a pre-built list of words that fall in each of these categories & then just implement a counting based approach(list is on web).
It surely cannot handle negation sentences like: “I am not feeling stressed or Depressed”. in which case it would give opposite results…

Further you may like to make your own corpus like 200-400 sentences, each one labelled with an emotion, and train a lstm, or GRU maybe, to see if it works…
You can take help of Google Word Embeddings, nltk-POS Tags as well.


#3

Thanks Rajat for your reply, coincidentally my response to our client was quite similar :slight_smile:
We have pretty similar algorithm currently working only on positive and negative sentiments.

Definitely it can be enriched with different corpus(classification of emotions). I agree with you on that. But not necessarily anxiety and stress could be expressed in different words, people use sometime same words for different emotions. I was just thinking if other alternative are possible. It seems like based on words itself seems to be more convincing, I will research more on it.

Thanks,
Shubham