Table 4. Ablation experiment results on iFlytek.
| Dataset (iFlytek) | Validation | |||
|---|---|---|---|---|
| Evaluation index (%) | Acc | P | R | F1 |
| Chinese-BERT-wwm-GCN | 60.50 | 59.21 | 60.50 | 59.90 |
| Chinese-BERT-wwm-GCN-L | 60.79 | 60.21 | 60.79 | 60.50 |
| Chinese-BERT-wwm-GCN-P | 60.91 | 59.83 | 60.91 | 60.37 |
| Chinese-BERT-wwm-GCN-LP | 61.95 | 60.27 | 61.95 | 61.01 |
| Chinese-RoBERTa-wwm-GCN | 60.79 | 59.19 | 60.79 | 59.98 |
| Chinese-RoBERTa-wwm-GCN-L | 61.13 | 60.37 | 61.13 | 60.75 |
| Chinese-RoBERTa-wwm-GCN-P | 61.09 | 60.30 | 61.09 | 60.69 |
| Chinese-RoBERTa-wwm-GCN-LP | 61.25 | 61.58 | 61.25 | 61.41 |
Note:
Remove one or two modules from our model to verify the effect of each module. We run all models three times and report the mean result on validation and test sets.