Detection of interaction articles and experimental methods in biomedical literature

BMC Bioinformatics

Table 2 Results and properties of our official runs.

Run	Acc	Spec	Sens	F-Score	MCC	AUC iP/R	ME	Feat	DTH
Official run 1	88.68	97.64	38.57	50.83	0.48297	63.85	+	WMP	0.50
Official run 2	87.93	93.06	59.23	59.82	0.52727	63.89	+	WMP	0.20
Official run 3	67.05	64.19	83.08	43.34	0.34244	41.74	–	P
Official run 4	73.68	74.13	71.21	45.08	0.34650	41.74	–	P
Official run 5	88.00	94.40	52.20	56.89	0.50255	62.39	+	WM	0.25
Post hoc run 2a	86.90	90.57	66.37	60.58	0.53089	64.06	+	WMP	0.20
Post hoc run 6	87.53	91.57	64.95	61.24	0.53969	66.30	+	WMPBS	0.21

Additionally we give our post hoc run 2a, performed without a minimal feature count threshold of 3 and bi-normal feature selection (20,000). Our best post hoc run 6 uses bigrams (B) and syntactic features (S). Features considered are W (bag of words), M (MeSH), P (PPIscore), B (bigrams), S (syntactic) - for a detailed description see page 3.

ISSN: 1471-2105