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

Table 6

Weighted average of different named entity recognition models.

No	Models	(%)	(%)	(%)
I	HMM	93.04	92.02	92.33
II	CRF	93.42	93.57	93.42
III	BiLSTM	93.57	93.65	93.48
IV	BiLSTM-CRF	94.29	94.42	94.22
V	(BiLSTM-CNN)SPI-CRF	94.59	94.67	94.49
VI	BiLSTM-CNN-CRF	94.82	94.88	94.71
VII	(BiLSTM-CNN)SPI-MHA-CRF	95.13	95.15	94.94
VIII	BiLSTM-CNN-MHA-CRF	95.39	95.48	95.37