Twitter sentiment analysis with BERT

This repo contains resources to further pre-train and fine-tune the BERT model for twitter sentiment analysis. The code was written/adapted prior to tensorflow team lauching the "classify text with bert" guide. If you'd like a simpler method for training the BERT model for TSA, I'd recommend you to follow the mentioned guide. I plan to implement this code according to such, in the near future.

BERT vs other approaches

Model	Avg Recall	F1	Accuracy
BB_twtr	0.681	0.685	0.658
DataStories	0.681	0.677	0.651
VADER	0.524	0.567	0.530
BERT_M4	0.711	0.703	0.697