Table 5 A comparison of four templates on four datasets
From: Aspect-level sentiment classification with fused local and global context
Template | Models | Restaurant | Laptop | MAMS | |
|---|---|---|---|---|---|
Accuracy | Accuracy | Accuracy | Accuracy | ||
The sentence is '{sentence}', where {aspect} means [Mask] | PConvBERT | 86.96 | 81.66 | 76.73 | 84.36 |
PConvRoBERTa | 89.29 | 83.54 | 78.47 | 84.96 | |
The sentence is '{sentence}', where I felt the {aspect} was [Mask] | PConvBERT | 86.43 | 78.21 | 75.00 | 83.08 |
PConvRoBERTa | 87.05 | 81.66 | 76.45 | 85.10 | |
The sentence is '{sentence}', where the {aspect} made me feel [Mask] | PConvBERT | 85.80 | 80.88 | 73.99 | 83.08 |
PConvRoBERTa | 86.52 | 82.13 | 75.29 | 84.21 | |
The sentence is '{sentence}', where the {aspect} is [Mask] | PConvBERT | 85.62 | 79.15 | 75.14 | 83.98 |
PConvRoBERTa | 88.39 | 82.45 | 74.86 | 85.55 | |
– | PConvBERT | 85.36 | 76.96 | 75.43 | 83.08 |
PConvRoBERTa | 86.43 | 83.07 | 73.84 | 84.88 |