Table 4.
Accuracy, precision, recall, and F1-score of Skip-gram and TopicS with different classification models.
| Model (Embed_sizea) | Accuracy (%) | Precision (%) | Recall (%) | F1-score (%) | Time (s) | |||||
| TextCNN | ||||||||||
|
|
Skip-gram (150 dimens) | 94.85 | 95.32 | 94.50 | 94.88 | 40.30 | ||||
|
|
TopicS (150 dimens) | 96.10b | 95.94 | 96.30 | 96.10 | 35.23 | ||||
| TextRNN | ||||||||||
|
|
Skip-gram (150 dimens) | 94.85 | 95.32 | 94.50 | 94.88 | 40.30 | ||||
|
|
TopicS (150 dimens) | 96.10b | 95.94 | 96.30 | 96.10 | 35.23 | ||||
| Transformer | ||||||||||
|
|
Skip-gram (100 dimens) | 85.45 | 85.06 | 78.80 | 81.13 | 55.16 | ||||
|
|
TopicS (150 dimens) | 90.70 | 90.90 | 90.60 | 90.68 | 49.32 | ||||
aEmbed_size represents the word-embedding size.
bItalicization represents that the metrics of TopicS are better than Skip-gram for each model.