Motivation: Protein–protein interactions (PPIs) play an important role in understanding biological processes. Although recent research in text mining has achieved a significant progress in automatic PPI extraction from literature, performance of existing systems still needs to be improved. Results: In this study, we propose a novel algorithm for extracting PPIs from literature which consists of two phases. First, we automatically categorize the data into subsets based on its semantic properties and extract candidate PPI pairs from these subsets. Second, we apply support vector machines (SVMs) to classify candidate PPI pairs using features specific for each subset. We obtain promising results on five benchmark datasets: AIMed, BioInfer, HPRD50, IEPA and LLL with F-scores ranging from 60% to 84%, which are comparable with the state-of-the-art PPI extraction systems. Furthermore, our system achieves the best performance on cross-corpora evaluation and comparative performance in terms of computational efficiency. Availability: The source code and scripts used in this article are available for academic use at http://staff.science.uva.nl/∼bui/PPIs.zip