How were these stocktwits data being labelled?
Hi!
Love your CryptoBERT, but I am wondering how these stocktwits that used to fine-tune CryptoBERT were being labelled?
Is there any crypto-jargon being added manually? Or CryptoBERT is finetunned with weak supervision?
Thank you so much!
Hi, thanks for the question.
The posts were labelled by their authors as either Bullish or bearish. If no label was given, we assumed the post to be neutral. These labels were passed to our model, when fine-tuning CryptoBERT on the sentiment classification task.
On a side note, the research should soon be published in the next edition of the IEEE Intelligent Systems Journal. So for more information, feel free to check out the article. The title is "Sentiment Classification of Cryptocurrency-related Social Media Posts".
Cheers!
Fantastic!
I am also working on my dissertation with similar topic for my master degree, can't wait to learn from your paper.
Well done and wish you the best :) :)!