Research on Named Entity Recognition Method of Metro On-Board Equipment Based on Multiheaded Self-Attention Mechanism and CNN-BiLSTM-CRF

Table 4

Recognition effect of different named entity recognition models on measure.

Models	B-measure			M-measure			E-measure
	P (%)	R (%)	F1 (%)	P (%)	R (%)	F1 (%)	P (%)	R (%)	F1 (%)
HMM	67.14	85.42	75.20	77.99	94.88	85.61	75.71	96.36	84.80
CRF	84.62	80.00	82.24	89.40	76.38	82.38	90.38	85.45	87.85
BiLSTM	86.00	78.18	81.90	92.11	68.90	78.83	91.67	80.00	85.44
BiLSTM-CRF	92.59	84.75	88.50	91.37	82.04	86.46	89.29	84.75	86.96
BiLSTM-CNN-CRF	92.41	84.94	88.52	90.76	79.58	84.80	94.34	84.75	89.29
BiLSTM-CNN-MHA-CRF	90.74	83.50	86.73	96.23	86.44	91.07	89.58	81.69	85.45