Semantic role labeling for protein transport predicates

BMC Bioinformatics

Table 4 Model performance by predicate type.

	Feature Set	Precision	Recall	F-measure	% of Roles
Nominal predicates	Phrase-Chunking	78.8	63.0	70.7	79.2
Verbal predicates	Phrase-Chunking	64.3	52.9	58.1	20.8
Nominal predicates	Protein-Transport	86.2	75.8	80.6	79.2
Verbal predicates	Protein-Transport	88.5	67.6	76.7	20.8
Wall Street Journal predicates	Phrase-Chunking	74.0	62.7	67.9	34.3
GeneRIF-only predicates	Phrase-Chunking	77.3	60.7	68.0	65.7
Wall Street Journal predicates	Protein-Transport	87.0	79.7	83.2	34.3
GeneRIF-only predicates	Protein-Transport	86.3	72.1	78.6	65.7

This table shows precision, recall, F-measure split up by the predicate part of speech and domain. A predicate was considered to be in the Wall Street Journal domain if the word appeared anywhere in the Wall Street Journal section of the Penn TreeBank, and was considered a GeneRIF-only predicate otherwise. The second column indicates which model (which feature set) the performance numbers are for. The final column indicates the percent of roles accounted for by each part of speech and each predicate domain.

ISSN: 1471-2105