Hits:

Indexed by:期刊论文

Date of Publication:2019-01-01

Journal:INTERNATIONAL JOURNAL OF DATA MINING AND BIOINFORMATICS

Included Journals:SCIE

Volume:22

Issue:2

Page Number:131-148

ISSN No.:1748-5673

Key Words:PPI article classification; self-attention; convolutional neural network; domain knowledge

Abstract:Automatic classification of protein-protein interaction (PPI) relevant articles from biomedical literature is a crucial step for biological database curation since it can help reduce the curation burden at the initial stage. However, most popular PPI article classification methods are based on traditional machine learning and their performances are heavily dependent on the feature engineering. Recent years, PPI article classification with neural networks has gained increasing attention, but domain knowledge has been rarely used in these methods. Aiming to exploit domain knowledge, we propose a domain Knowledge-enriched Self-Attention Convolutional Neural Network (KeSACNN) approach for PPI article classification. In this approach, two knowledge embeddings are proposed, and the novel convolution neural network architectures with self-attention mechanism are designed to leverage biomedical knowledge. The experimental results show that our method achieves the state-of-the-art performance on the BioCreative II and III corpora (82.92% and 67.93% in F-scores, respectively).