Detection of interaction articles and experimental methods in biomedical literature

BMC Bioinformatics

Table 3 Comparison of mean results of a stratified 10-fold cross-validation experiment on the development set using oversampling versus using discretization threshold (DTH) lowering.

Method	Acc	Spec	Sens	F-Score	MCC	AUC iP/R	Feat	DTH
Oversampling	86.72	92.22	59.93	60.49	0.52580	66.36	WMP	0.50
DTH lowering	85.94	89.76	67.35	61.92	0.53689	69.33	WMP	0.20
DTH lowering	87.84	93.81	58.82	62.21	0.55363	69.33	WMP	0.36

No restrictions on minimal feature occurrence or feature set size were used in these experiments, therefore the DTH of 0.20 is no longer optimal. A DTH of 0.36 gives the best results overall. 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