Table 10 Ablation studies of hyperparameter on CH-SIMS and CMU-MOSI dataset.
From: Multimodal sentiment analysis based on multi-layer feature fusion and multi-task learning
Model design | SIMS | MOSI | ||
|---|---|---|---|---|
Acc2 | Acc5 | Acc2 | Acc7 | |
(1) Learning rate | Â | Â | Â | Â |
1e−3/5e−3/1e−4 | 78.36/80.56/79.60 | 42.54/45.20/45.00 | 82.4/86.6/86.1 | 42.4/46.7/46.3 |
(2) Batch size | Â | Â | Â | Â |
16/32/64 | 80.56/78.74/74.39 | 45.20/41.55/36.27 | 86.2/86.6/83.2 | 46.2/46.7/42.1 |
(3) Epoch | Â | Â | Â | Â |
30/50/100 | 77.29/80.56/79.83 | 43.35/45.20/44.83 | 86.6/86.3/86.3 | 46.7/46.4/46.4 |
(4) Early stop | Â | Â | Â | Â |
4/8/10 | 80.56/80.56/78.91 | 45.20/45.20/43.91 | 86.6/86.6/85.9 | 46.7/46.7/45.9 |
