Table 5 Ablation study of the proposed GE-Net, LA-Net, and IE Gate (%).
From: Adaptive feature interaction enhancement network for text classification
Model name | Group | GE | FLA | Gate | F1 |
|---|---|---|---|---|---|
TextCNN | Group A | \(\checkmark\) | \(\checkmark\) | \(\checkmark\) | 97.55 |
A1 | \(\checkmark\) | – | – | 91.03 | |
A2 | \(\checkmark\) | \(\checkmark\) | - | 90.81 | |
DPCNN | Group B | \(\checkmark\) | \(\checkmark\) | \(\checkmark\) | 93.62 |
B1 | \(\checkmark\) | – | – | 91.12 | |
B2 | \(\checkmark\) | \(\checkmark\) | – | 90.97 | |
LSTM (3) | Group C | \(\checkmark\) | \(\checkmark\) | \(\checkmark\) | 91.56 |
C1 | \(\checkmark\) | – | – | 91.00 | |
C2 | \(\checkmark\) | \(\checkmark\) | – | 90.69 | |
Transformer | Group D | \(\checkmark\) | \(\checkmark\) | \(\checkmark\) | 90.01 |
D1 | \(\checkmark\) | – | – | 89.74 | |
D2 | \(\checkmark\) | \(\checkmark\) | – | 89.80 |
