Swift.ai: research and development of an integrated platform for machine-assisted research synthesis

点击翻译按钮获取中文摘要

Project Abstract (30 lines of text) 1 Systematic review and evidence mapping, both forms of research synthesis, are formal, sequential processes 2 for identifying, assessing, and integrating the primary scientific literature. These approaches, already 3 cornerstones of evidence-based medicine, have recently gained significant popularity in several other 4 disciplines including environmental, agricultural, and public health research and are increasingly utilized for 5 informed decision making by governmental organizations. It has been estimated that more than 25,000 6 systematic reviews are conducted and published annually and selecting studies for inclusion is one of the most 7 resource intensive steps for any systematic review or evidence map. In Phase I of our research plan, we have 8 developed a web-based, collaborative systematic review web application called SWIFT-Active Screener, an 9 innovative document screening tool that allows users to identify the majority of relevant articles after screening 10 only a fraction of the total number of abstracts. Our goal for the current proposal is to conduct additional 11 research and development required to make SWIFT-Active Screener a commercial success, while also 12 building on and leveraging methods and software we have previously built to address other stages in the 13 systematic review pipeline. Therefore, one of the primary aims of our ongoing research and development is to 14 address this need by expanding the Active Screener application into an integrated platform for research 15 synthesis by uniting it with several of our other related software products. The resulting platform, which we call 16 “swift.ai,” is described in detail in “Aim 1 – Software engineering to create a unified platform for research 17 synthesis.” In “Aim 2 – Improved statistical methods for Active Screener 2.0”, we expand on the methodological 18 research completed during Phase I of this SBIR, to further develop and refine our methods. Specifically, we 19 investigate new ways to integrate state-of-the art methods in deep learning and new ways to better utilize the 20 large amounts of screening data collected from our users in order to improve our models. Finally, in “Aim 3 – 21 Living evidence maps powered by Active Screener 2.0,” we explore new approaches for using machine 22 learning to facilitate evidence mapping.