System for Semi-Automated Literature Review Based on Machine Learning

Abstract

This paper presents the design and implementation of a system for semi-automating the literature review process based on machine learning. By using machine learning algorithms, the system determines whether scientific papers belong to the topic that is being explored as part of the review process. The system’s user interface allows the process of creating a literature review to be managed through a series of steps: selecting data sources, building queries and topic searches, displaying the scientific papers found, selecting papers that belong to the set of desired papers, running machine learning algorithms for learning and automated classification, and displaying and exporting the final set of papers. Manual literature reviews are compared with automated reviews, and similarities and differences between the two approaches in terms of duration, accuracy, and ease of use are discussed. This study concludes that the best results in terms of sensitivity and accuracy for the automated literature review process are achieved by using a combined machine learning model, which uses multiple unweighted machine learning models. Cross-testing the models on two alternative datasets revealed an overlap in the machine learning hyperparameters. The stable sensitivity and accuracy in the tests indicate the potential for generalized use in automated literature review.