Towards the automation of systematic reviews using natural language processing, machine learning, and deep learning: a comprehensive review

Abstract

Systematic reviews (SRs) constitute a critical foundation for evidence-based decision-making and policy formulation across various disciplines, particularly in healthcare and beyond. However, the inherently rigorous and structured nature of the SR process renders it laborious for human reviewers. Moreover, the exponential growth in daily published literature exacerbates the challenge, as SRs risk missing out on incorporating recent studies that could potentially influence research outcomes. This pressing need to streamline and enhance the efficiency of SRs has prompted significant interest in leveraging Artificial Intelligence (AI) techniques to automate various stages of the SR process. This review paper provides a comprehensive overview of the current AI methods employed for SR automation, a subject area that has not been exhaustively covered in previous literature. Through an extensive analysis of 52 related works and an original online survey, the primary AI techniques and their applications in automating key SR stages, such as search, screening, data extraction, and risk of bias assessment, are identified. The survey results offer practical insights into the current practices, experiences, opinions, and expectations of SR practitioners and researchers regarding future SR automation. Synthesis of the literature review and survey findings highlights gaps and challenges in the current landscape of SR automation using AI techniques. Based on these insights, potential future directions are discussed. This review aims to equip researchers and practitioners with a foundational understanding of the basic concepts, primary methodologies, and recent advancements in AI-driven SR automation while guiding computer scientists in exploring novel techniques to invigorate further and advance this field.