Semantic role labeling for protein transport predicates

BMC Bioinformatics

Table 3 Model performance by seen vs. unseen.

	Precision	Recall	F-measure	%
Seen roles	97.7	88.9	93.1	60.1
Unseen roles	71.6	63.6	67.4	39.2
Seen AGENT roles	100.0	100.0	100.0	33.3
Unseen AGENT roles	100.0	0.0	0.0	66.7
Seen PATIENT roles	100.0	86.7	92.9	37.7
Unseen PATIENT roles	78.3	66.9	72.2	62.3
Seen ORIGIN roles	97.1	84.6	90.4	86.7
Unseen ORIGIN roles	14.3	16.7	15.4	13.3
Seen DESTINATION roles	96.5	91.6	94.0	86.2
Unseen DESTINATION roles	57.1	63.2	60.0	13.8

This table shows precision, recall, F-measure comparing role phrases which appeared in both the training and testing data ("seen" roles) to role phrases which appeared only in the testing data ("unseen" roles). Both overall results and results by role type are shown. The final column indicates the percent of the particular category of roles that were seen (or unseen). Performance numbers were calculated from the output of the Protein-Transport model when using manually annotated protein boundaries.

ISSN: 1471-2105