# Simulation-based Active Learning for Systematic Reviews: A Systematic Review of the Literature - Repository Welcome to the repository for the academic paper, "Simulation-based Active Learning for Systematic Reviews: A Systematic Review of the Literature" This repository serves as a centralized storage for all pertinent files, results, and resources related to our systematic investigation of active learning's potential to improve the efficiency and effectiveness of the screening phase in systematic reviews. Our research aims to consolidate the fragmented field of active learning research and provide a unified understanding of the benefits and limitations of active learning in the context of systematic reviews. In this repository, you will find the following materials: - Raw data from the literature search - Project files from the literature screening - Result files - Extracted metrics ## Background Active learning (or: TAR - Technology Assisted Review, CAL - Continuous Active Learning) has emerged as a promising approach to improve the efficiency and effectiveness of the screening phase in systematic reviews. The exploration of active learning solutions in systematic reviews has been extensive, encompassing both theoretical research and simulation studies. However, the field of research is currently fragmented, resulting in an environment where the benefits of active learning have not been fully realized. To help conventionalize active learning as the solution for systematic review automation and help inform researchers on the status of active learning review research, this study collects and reports on literature related to active learning reviews. The aim is to answer whether active learning is to be recommended for use in systematic review acceleration. Furthermore, the study is intended to benefit future research on active learning by identifying which areas have been well-studied and which require further investigation. ## Method We conducted a systematic review to investigate the use of active learning in systematic reviews. We searched multiple databases for literature on the topic published after 2005 and used the open-source software ASReview with an active learning algorithm to screen the abstracts for relevance. The resulting datasets were screened for relevant literature. From these documents information is extracted. We collect information on study set-up and design, on dataset availability and usage, and how active learning is used. ## Results Out of 1548 articles, 48 were labeled as relevant to the study. These articles focus on the use of active learning to improve the efficiency of the screening phase in systematic reviews. Many variables were extracted, and are presented in tables. On top of the 48 identified studies, 305 datasets were collected, of which 208 were incorporated into the review. ## Discussion Our systematic review highlights the potential of active learning as a means to accelerate the screening phase in systematic reviews, with every single analyzed paper recommending its use. Despite the identified limitations, the consistent recommendation for active learning showcases its promise in improving systematic review automation. To further advance the field, future research should focus on standardizing metrics, promoting open data practices, and exploring a wider variety of models, ultimately benefiting both researchers and practitioners in the systematic review domain.