Research Article
Ptr4BERT: Automatic Semisupervised Chinese Government Message Text Classification Method Based on Transformer-Based Pointer Generator Network
Table 2
Performance comparison of different methods.
| Method | HNMes | QDMes | P | R | F1 | P | R | F1 |
| Logistic regression | 72.11 | 71.70 | 71.91 | 68.69 | 68.15 | 68.41 | Naïve Bayes | 81.42 | 79.46 | 80.43 | 77.95 | 78.86 | 78.40 | SVM | 84.67 | 82.19 | 83.41 | 86.54 | 84.39 | 85.45 | TextRNN | 90.02 | 90.06 | 90.04 | 89.53 | 88.82 | 89.17 | TextCNN | 90.32 | 89.27 | 89.79 | 88.07 | 88.81 | 88.93 | TextRCNN | 90.14 | 90.07 | 90.10 | 89.26 | 89.19 | 89.22 | BERT | 93.87 | 93.14 | 93.50 | 92.18 | 91.62 | 91.89 | ERNIE | 93.52 | 92.71 | 93.11 | 92.76 | 92.01 | 92.38 | Ptr4BERT (ours) | 94.39 | 94.38 | 94.39 | 93.36 | 93.39 | 93.37 |
|
|