Table 3.

Word-embedding dimension parameters with TextCNN.

Evaluation metrics and model			100		150		200	250			300		350		400
Accuracy (%)
	Skip-gram	93.75		94.85		94.00			93.15	93.80		93.49		93.40
	TopicSa	95.10b		96.10		94.80			94.45	94.95		95.25		95.40
Precision (%)
	Skip-gram	91.92		95.32		94.07			93.16	93.33		93.28		93.09
	TopicS	96.32		95.95		95.56			94.11	95.20		95.87		96.53
Recall (%)
	Skip-gram	94.00		94.50		94.00			93.30	94.40		93.90		93.90
	TopicS	95.90		96.30		94.00			94.90	94.70		94.60		94.20
F1-score (%)
	Skip-gram	93.90		94.88		93.95			93.17	93.84		93.48		93.44
	TopicS	95.09		96.10		94.77			94.48	94.94		95.22		95.34

^aTopicS represents the topic-enhanced word-embedding model proposed in this paper.

^bItalicization represents that the metrics of TopicS are better than Skip-gram for each metric.