Table 5.
Performance measures of various text models
| Text model | Validation accuracy | Test accuracy |
|---|---|---|
| BERT (Baseline) | 86.54% | 86.44% |
| InferSent (Baseline) | 86.34% | 86.31% |
| LSTM+CNN | 85.51% | 85.25% |
| BiGRU+Capsule | 85.98% | 86.18% |
| BiLSTM+BiGRU+attention | 87.89% | 87.90% |
| 2D CNN | 86.52% | 86.77% |
| BERT+Dense | 89.34% | 89.46% |
| RoBERTa+Dense | 88.52% | 88.62% |
Bold indicates models with better performance measures (here validation and Test accuracy)