POS-tags features for Protein-Protein Interaction Extraction from Biomedical Articles

Abstract

Protein-Protein Interaction (PPI) extraction from biomedical articles consists on extracting sentences were two or more proteins interact. Traditional articles tackle this problem creating more sophisticated classifiers. In contrast to them, we focus on discriminative features that can be exploited by traditional classifiers. Our method exploits information from POS-tags features and are combined with a bag-of-words approach. We used five standard corpora of PPI: Aimed, Bioinfer, HPRD50, IEPA and LLL. Our method is simple and achieves high results compared with other approaches. We achieve an improvement of 11% with our best competitor.