Classifying protein-protein interaction articles using word and syntactic features

BMC Bioinformatics

Table 6 Performance changes on the ACT development set by varying feature types.

Used Features	Avg Prec	Precision	Recall	F1 score
Baseline	0.7073	0.6403	0.6290	0.6346
–Gene Anonymization	0.7017	0.6166	0.6320	0.6242
–Multi-words	0.7035	0.6358	0.6349	0.6354
–Sub-strings	0.7019	0.6329	0.6320	0.6324
–MeSH Terms	0.7009	0.6410	0.6334	0.6372
Baseline+Higher Order	0.7077	0.6311	0.6496	0.6402

The baseline performance is the result obtained from our system pipeline with the same setting used for Run 4. A row shows the evaluation results when a specific feature type is not used for the experiment. However, the last row is the performance results when higher-order features are applied.

ISSN: 1471-2105