A Chinese Named Entity Recognition Model of Maintenance Records for Power Primary Equipment Based on Progressive Multitype Feature Fusion

Table 2

Performance comparison of different modules on EPE-MR dataset (CE: character embedding; RFR: radical feature representation).

Modules	Metrics
	P	R	F1
LSTM	0.815	0.707	0.741
BiLSTM	0.811	0.776	0.784
CE	0.774	0.743	0.762
CE-CRF	0.828	0.769	0.795
BiLSTM-CRF	0.834	0.805	0.807
CE-BiLSTM-CRF	0.869	0.866	0.862
CE + RFR-BiLSTM-CRF (the proposed method)	0.892	0.886	0.883

Bold values show the best performance.