Asian Journal of Complementary and Alternative Medicine

Short Review on srBERT: Automatic Article Classification Model for Systematic Review Using BERT

Seon Choe¹, Sungmin Aum^2,3, Ju Han Kim¹*

¹Seoul National University Biomedical Informatics (SNUBI), Division of Biomedical Informatics, Seoul National University College of Medicine, Seoul, 03080, South Korea 

²Korea Institute of Science and Technology (KIST), Robotics and Media Institute 5, Hwarang-ro 14-gil, Seongbuk-gu, Seoul

³University of Science and Technology (UST), Division of Nano and Information Technology, 34113, Gajeong-ro, Yuseong-gu, Daejeon, Republic of Korea  

Abstract

Systematic reviews (SRs) have been recognized as the most rigorous and reliable approach to enable evidence-based medicine. However, the considerable workload required to create SRs prevents reflecting the latest knowledge. This study automated the classification of included articles by adopting the Bidirectional Encoder Representations from Transformers (BERT) algorithm. By pretraining with abstracts of articles and a generated vocabulary fine-tuned with titles of articles, the proposed srBERT_my overcomes the training data insufficiency while improving performance in both classification and relation-extraction tasks. Despite the limitation of model vulnerabilities based on training dataset quality, the results demonstrated the feasibility of automatic article classification using machine-learning (ML) approaches to support SR tasks

Keywords:

Systematic review, process automation, deep learning, text mining