I’m trying to create an NLP Emotion Classification Model for a research project but kind of confused on where and how to start. I have this huge dataset of Reddit posts and want to classify each post into like 12 different emotion categories.
Is there a way to do this using existing models eg. BERT or can I also do this using unsupervised learning?
I have at least 12000 different posts and so want to avoid supervised learning because its going to take so long to label a set for training data also I might lose a lot of time doing that.
Whats the most efficient and accurate way to do this? Any help would be amazing!
You must log in or register to comment.
You can label a balanced subset and try something like SeTFiT
Go google the setfit library on GitHub. Frame it as a few shot learning task, won’t get perfect results but seems tractable as a problem.
I doubt you’ll find 12 different emotions on Reddit. I think everything can fit into:
- sarcasm
- rage
- polemic rage
- bewilderment
- jocularity
- serious
I might have missed one or two, but I’m sure there aren’t 12.
Try the Go Emotions dataset from Google