A small project on sentiment classification of passenger's tweets
Dataset Twitter US Airline Sentiment
Which was tried:
- SVM over TF-IDF as baseline (f1_score: macro 0.7271, micro 0.7987)
- Flair model (f1_score: macro 0.7374, micro 0.7947)
- ULMFIT (f1_score: macro 0.7454, micro 0.8139)
- BERT (f1_score: macro 0.7830, micro 0.8323)
- Universal Sentence Encoder (f1_score: macro 0.7835, micro 0.8357)