How Can No/Low Code Platforms Help End-Users Develop ML Applications? - A Systematic Review

Abstract

AbstractWith the increasing popularity of machine learning, the demand for users to develop machine learning applications has grown rapidly, which brings about a rising growth for end-user development (EUD) method research. Based on previous works, EUD can be roughly divided into two categories: methods with coding and without coding (i.e. no-code). In recent years, no-code and low-code methods have become the mainstream EUD methods that have been widely concerned by the education and business community, due to their low technical barriers. However, there lacks a comprehensive summary of existing research to answer some fundamental questions, such as: How can no/low-code platform help end-users develop ML applications? what are their effects, design methods, and user experience? This paper answers the above questions through a systematic literature review. Two experienced researchers carefully read, coded, analyzed, and checked the literature by using MAXQDA, the results showed:1. No-code or low-code tools can already support the development pipeline of ML applications that traditionally requires coding. 2. No-code or low-code methods are preferred by users. 3. For design purposes, the visual development method is the most commonly used form, especially in the field of education. 4. In terms of interactive experience, a few design principles were recognized from the reviewed pieces of literature, including the interactive process should provide a low threshold, high ceiling, and wide walls; the information architecture should meet the mental model of the novice user's cognitive process; the platform functions should support rapid prototyping, iterations, and timely feedback.KeywordsSystematic mapping studyEnd-user developmentMachine learning