Exploiting graph kernels for high performance biomedical relation extraction

Table 2 Performance measurements for chemical induced disease relation extraction

Method	Sent-Rel. only			Non-Sent-Rel. only			All relations
	P	R	F	P	R	F	P	R	F
SSTK with CP-Tree	43.1	73.7	54.4	36.9	14.2	20.5	42.5	56.0	48.3
PTK with LCT	42.2	75.3	54.1	30.5	40.1	34.6	39.5	64.8	49.0
APG with Dep. Graphs	54.7	80.6	65.1	47.8	43.8	45.7	53.2	69.7	60.3
ASM with Dep. Graphs	51.6	80.8	63.0	38.8	36.0	37.3	49.0	67.4	56.8
Hybrid (Prior art [21])	-	-	-	-	-	-	64.9	49.2	56.0
Hybrid + Rules (Prior art [21])	-	-	-	-	-	-	55.6	68.4	61.3

(Key: P,R,F denotes Precision, Recall and F1 score respectively. Sent-Rel. and Non-Sent-Rel. denotes sentence level relations and Non-Sentence level relations respectively. CP-Tree and LCT denote constituency parse tree and location centered tree. Dep. Graph denotes dependency graph. The best performance is highlighted in italicized font)

ISSN: 2041-1480