The purpose of this project is to do a sentiment analysis on tweets about Covid-19 vaccinations in 3 classes (0-neutral, 1-negative, 2-positive). To-do this, I experimented with 4 different types of models:
- Softmax Regression.
- Feed-Forward neural network.
- Bidirectional RNN neural network with LSTM & GRU cells (witout and with attention layer).
- BERT-Base-uncased.
| Model | Precision | Recall | F1 Score | Accuracy |
|---|---|---|---|---|
| Softmax Regression | 71 % | 71 % | 71 % | 71.12 % |
| Feed-Forward neural network | 58 % | 66 % | 61 % | 65.6 % |
| LSTM neural network | 70 % | 66 % | 67 % | 65.55 % |
| LSTM with attention layer | 69 % | 72 % | 70 % | 69.36 % |
| BERT-Base-uncased | 78 % | 71 % | 73 % | 71 % |