Table 7 Component removal effect on model’s overall performance.
From: A deep sentiment model combining ALBERT-driven context and EHO-optimized architecture
Method | Lap14 | Res14 | Res15 | Res16 | SST-5 | |||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
Acc | F | Acc | F | Acc | F | Acc | F | Acc | F | Acc | F | |
Proposed method | 85.9 | 83.2 | 83.1 | 81.5 | 89.6 | 84.4 | 85.7 | 86.1 | 95.7 | 87.0 | 53.5 | 53.0 |
Without ALBERT | 81.0 | 78.6 | 79.5 | 77.8 | 85.0 | 80.1 | 82.0 | 81.6 | 83.5 | 79.8 | 51.0 | 50.6 |
Without Bi-GRU | 81.5 | 79.1 | 80.3 | 78.5 | 85.8 | 81.0 | 83.2 | 82.5 | 84.0 | 80.6 | 52.5 | 52.0 |
Without Bi-LSTM | 81.7 | 79.4 | 78.9 | 77.2 | 84.6 | 79.6 | 81.8 | 81.0 | 83.0 | 79.2 | 50.5 | 50.1 |
Without EHO | 82.3 | 80.1 | 80.2 | 78.9 | 86.2 | 81.9 | 82.3 | 82.0 | 84.5 | 81.3 | 51.5 | 51.0 |
