点击次数：

论文类型：期刊论文

发表时间：2015-04-15

发表刊物：Journal of Computational Information Systems

收录刊物：EI、Scopus

卷号：11

期号：8

页面范围：2823-2832

ISSN号：15539105

摘要：Automated extraction of Protein-protein interaction (PPI) from biomedical literatures is a significant task. While various methods have been presented to deal with this matter, many efforts are still needed to be made to further improve the extraction performance. In this paper, we propose a composite-kernel-based approach which combines feature-based kernel and graph kernel. The feature-based kernel not only utilizes rich local context features and dependency syntactic features but also introduces active learning. Besides, we perform the standard ten-fold cross-validation (CV) evaluation and cross-corpus evaluation strategies across five PPI corpora. The results suggest that our PPI system outperforms other approaches on four out of the five corpora. We obtain a promising F-score of 66.4 and AUC score of 90.9 on AIMed corpus, and achieve a 74.4 F-score and a 92.1 AUC score on BioInfer corpus. In addition, our system gains a comparative performance on cross-corpus evaluation. ?, 2015, Binary Information Press. All right reserved.