Reproducibility and Data storage Checklist for Active Learning-Aided Systematic Reviews

Abstract

In the screening phase of a systematic review, screening prioritization via active learning effectively reduces the workload. However, the PRISMA guidelines are not sufficient for reporting the screening phase in a reproducible manner. Text screening with active learning is an iterative process, but the labeling decisions and the training of the active learning model can happen independently of each other in time. So it is not trivial to store the data from both events so that you can still know which iteration of the model was used for each labeling decision. Moreover, many iterations of the active learning model will be trained throughout the screening process, producing an enormous amount of data (think of many gigabytes or even terabytes of data), and machine learning models are continually becoming larger. Together this can add up to an undesirable amount of data when naively storing all the data produced at every iteration of the active learning pipeline. This article clarifies the steps in an active learning-aided screening process and what data is produced at every step. We show how this data can be stored efficiently in terms of size. Most notably, the data produced by the model is where we need to strike a balance between reproducibility and storage size. Finally, we created the RDAL-Checklist (Reproducibility and Data storage for Active Learning-aided systematic reviews – checklist) that helps users and creators of active learning software make their screening process reproducible.

The SAFE procedure: a practical stopping heuristic for active learning-based screening in systematic reviews and meta-analyses

“…The proposed procedure is meant to determine when to stop screening when applying active learning-aided screening while adhering to the PRISMA 2020 statement [29] and Open Science principles to ensure reproducibility and transparency for AI-aided output [24]. It is designed to be conservative and easily understood by non-experts and to enable finding a reasonable percentage of relevant records in the dataset rather than aiming for 100% [8,30].…”

The SAFE Procedure: A Practical Stopping Heuristic for Active Learning-Based Screening in Systematic Reviews and Meta-Analyses

“…The proposed procedure is meant to determine when to stop screening when applying active learningaided screening while adherering to the PRISMA 2020 statement (Page et al, 2021) and Open Science principles to ensure reproducibility and transparency for AI-aided output (Lombaers et al, 2023). It is designed to be conservative and easily understood by non-experts and to enable the finding of a reasonable percentage of actual relevant records in the dataset, rather than aiming for 100% (Bramer et al, 2018;Papaioannou et al, 2010).…”

Hacia revisiones de la literatura más eficientes potenciadas por inteligencia artificial

“…La aplicación del aprendizaje activo a la revisión sistemática se le conoce como RITL, del inglés Researcher-In-The-Loop 15 , y es utilizada por algunas aplicaciones enfocadas a resolver prioritariamente algunas cuestiones del manejo de información. Principalmente, esta metodología es utilizada en el proyecto ASReview que ofrece un ecosistema donde incluye detección asistida por IA, infraestructura para personalizar los modelos, así como opciones para ejecutar sistemas de simulación para beneficiar principalmente el screening 16 . Una herramienta similar es Rayyan 17 , una plataforma que ofrece un sistema que promete aprender de las decisiones que toman los usuarios para la selección del material, además de ofrecer portabilidad y un ambiente colaborativo.…”