MIT Press

Table 3

Development results for detecting speculation CUES: Averaged 10-fold cross-validation results for the cue classifiers on both the abstracts and full papers in the BioScope training data (BSA and BSP).

	Sentence Level			Token Level			Cue Level
Model	Prec	Rec	F₁	Prec	Rec	F₁	Prec	Rec	F₁
Baseline	91.07	87.21	89.07	91.61	81.85	86.42	90.49	81.16	85.57
WbW	95.01	88.03	91.37	95.29	82.78	88.58	94.65	82.26	88.02
Filtering	94.52	89.72	92.04	94.88	84.86	89.57	94.13	84.60	89.11

	Sentence Level			Token Level			Cue Level
Model	Prec	Rec	F₁	Prec	Rec	F₁	Prec	Rec	F₁
Baseline	91.07	87.21	89.07	91.61	81.85	86.42	90.49	81.16	85.57
WbW	95.01	88.03	91.37	95.29	82.78	88.58	94.65	82.26	88.02
Filtering	94.52	89.72	92.04	94.88	84.86	89.57	94.13	84.60	89.11