Evaluating a Machine Learning-Aided Software for Systematic Reviews: ASReview LAB

• In Event: Methodological Innovations for Conducting Research Syntheses

Abstract

The goal of systematic reviews is to synthesize existing research to answer research questions. Systematic reviews are comprised of two steps: 1) screen potentially relevant article abstracts and 2) screen the full text of those relevant abstracts for full relevance. With the increasing number of publications, systematic reviews are more time-, labor-, and cost-intensive. Therefore, researchers have developed machine learning-aided programs to help with the first step. ASReview LAB (van de Schoot et al., 2021) utilizes machine and active learning to screen article titles and/or abstracts for relevance. A simulation study is conducted to determine the optimal training data and settings’ combinations to best achieve the ideal recovery of relevant articles using ASReview LAB compared to traditional manual reviews.

Authors

• Ashani Jayasekera, University of Maryland

• Tracy Sweet, University of Maryland

• Laura M. Stapleton, University of Maryland

• Jason C. Chow, Vanderbilt University