Table 3 Performance comparison between our method and baselines on MOSI Dataset.
From: Multimodal sentiment analysis based on multi-layer feature fusion and multi-task learning
Models | MOSI | Â | |||
|---|---|---|---|---|---|
Acc2 | F1-score | MAE | Corr | Acc7 | |
EF-\(\hbox {LSTM}^\textrm{a}\) | 77.8/79.0 | 77.7/78.9 | 0.952 | 0.651 | 34.5 |
LF-\(\hbox {DNN}^\textrm{a}\) | 78.0/79.3 | 77.9/79.3 | 0.978 | 0.658 | 33.6 |
\(\hbox {TFN}^\textrm{b}\) | -/80.8 | -/80.7 | 0.901 | 0.698 | 32.1 |
\(\hbox {LMF}^\textrm{b}\) | -/82.5 | -/82.4 | 0.917 | 0.695 | 32.8 |
\(\hbox {MFN}^\textrm{b}\) | 77.4/- | 77.3/- | 0.965 | 0.632 | 34.1 |
Graph-\(\hbox {MFN}^\textrm{a}\) | 77.9/80.2 | 77.8/80.1 | 0.939 | 0.656 | 34.4 |
\(\hbox {MulT}^\textrm{b}\) | 81.5/84.1 | 80.6/83.9 | 0.861 | 0.711 | 40.0 |
\(\hbox {MISA}^\textrm{b}\) | 81.8/83.4 | 81.7/83.6 | 0.783 | 0.761 | 42.3 |
Self-\(\hbox {MM}^*\) | 82.5/83.8 | 82.5/83.9 | 0.758 | 0.779 | 44.9 |
\(\hbox {MMIM}^*\) | 80.2/81.9 | 80.2/82.0 | 0.811 | 0.725 | 45.0 |
Our Method | 85.2/86.6 | 85.2/86.7 | 0.728 | 0.792 | 46.7 |
