A Topic Recognition Method of News Text Based on Word Embedding Enhancement

Table 4

Results of the 20NewsGroup in 20 classes for 7532 texts by SVM and LR.

Model	Average 5-fold micro-F1 score of different dimensions
	100			200			300			400			500
		SVM	LR		SVM	LR		SVM	LR		SVM	LR		SVM	LR
TF-IDF	100	0.373	0.372	200	0.48	0.475	300	0.586	0.580	400	0.637	0.631	500	0.671	0.663
LDA	100	0.686	0.682	200	0.692	0.689	300	0.712	0.711	400	0.715	0.714	500	0.721	0.723
Glove	100	0.724	0.710	200	0.767	0.754	300	0.784	0.771	400	0.794	0.780	500	0.799	0.788
SGL	100	0.745	0.732	200	0.779	0.767	300	0.792	0.780	400	0.807	0.794	500	0.813	0.802
CGL	200	0.782	0.772	400	0.804	0.792	600	0.812	0.796	800	0.822	0.806	1000	0.826	0.813
Word2vec	100	0.740	0.733	200	0.765	0.756	300	0.766	0.755	400	0.769	0.758	500	0.765	0.755
SWL	100	0.747	0.743	200	0.777	0.769	300	0.782	0.771	400	0.787	0.777	500	0.787	0.777
CWL	200	0.780	0.776	400	0.795	0.787	600	0.793	0.782	800	0.793	0.783	1000	0.796	0.785

Bold indicates that values are the optimal results.