srBERT: automatic article classification model for systematic review using BERT.

Area of Science:

• Natural Language Processing
• Machine Learning
• Evidence-Based Medicine

Background:

• Systematic reviews (SRs) are vital for evidence-based medicine but are time-consuming.
• Automation tools for SRs face challenges with domain-specific data and limited training text.
• Existing natural language processing models struggle with specialized scientific literature.

Purpose of the Study:

• To automate the classification of articles for systematic reviews using a Bidirectional Encoder Representations from Transformers (BERT) algorithm.
• To develop and evaluate srBERT models pre-trained on article abstracts and fine-tuned on titles.
• To compare the performance of proposed srBERT models against general machine learning models.

Main Methods:

• Utilized Bidirectional Encoder Representations from Transformers (BERT) for article classification.
• Developed srBERT models pre-trained on abstracts from two datasets and fine-tuned using article titles.
• Compared srBERT models with existing general machine learning approaches.

Main Results:

• The proposed srBERT_my model achieved state-of-the-art performance in classification and relation-extraction tasks.
• Achieved 94.35% accuracy and 66.12 F1 score in classification on the original dataset.
• Demonstrated superior performance over the original BERT model in classification and relation extraction.

Conclusions:

• Machine learning, specifically srBERT, shows promise for automating article classification in systematic reviews.
• The model's performance is dependent on the quality and characteristics of the training dataset.
• Ensuring high-quality, sufficiently sized datasets is crucial for successful implementation.