Table 7.

The performance of biomedical event extraction on the BioNLP shared task 2011 Genia event corpus.

Method and event type		Precision (%)	Recall (%)	F1 score (%)
TEESa,b
	Event totalc	57.65	49.56	53.30
EventMinea
	Event total	63.48	53.35	57.98
Stacked generalizationa
	Event total	66.46	48.96	56.38
TEES-CNNsa,d
	Event total	69.45	49.94	58.07
HANNe,f
	Event total	71.73	53.21	61.10
KBg-driven tree LSTMe,h
	Simple totali	85.95	72.62	78.73
	Binding	53.16	37.68	44.10
	Regulation totalj	55.73	41.73	47.72
	Event total	67.10	52.14	58.65
GEANet-SciBERTe,k
	Regulation total	55.21	47.23	50.91
	Event total	64.61	56.11	60.06
DeepEventMinee
	Regulation total	62.36	51.88	56.64l
	Event total	76.28	55.06	63.96l
Our modele
	Simple total	82.23	78.88	80.52
	Binding	55.12	37.48	44.62
	Regulation total	57.82	46.39	51.48
	Event total	72.62	53.33	61.50

^aPipeline model.

^bTEES: Turku Event Extraction System.

^cRepresents the overall performance on the test set.

^dCNN: convolutional neural network.

^eJoint model.

^fHANN: hierarchical artificial neural network.

^gKB: knowledge base.

^hLSTM: long short-term memory.

ⁱRepresents the overall performance for simple events on the test set.

^jRepresents the overall performance for nested events on the test set (including regulation, positive regulation, and negative regulation subevents).

^kGEANet-SciBERT: Graph Edge-conditioned Attention Networks with Science BERT.

^lThe best value compared with other models.