A Deep Learning Model for the Assessment of the Visual Aesthetics of Mobile User Interfaces

Abstract

Visual aesthetics is one of the first aspects that users experience when looking at graphical user interfaces (GUIs), contributing to the perceived usability and credibility of a software system. It can also be an essential success factor in contexts where graphical elements play an important role in attracting users, such as choosing a mobile app from an app store. Therefore, visual aesthetics assessments are crucial in interface design, but traditional methods, involving target user representatives assessing each GUI individually, are costly and time-consuming. In this context, machine learning models have been demonstrated to be promising in automating the assessment of GUIs of web-based software systems. Yet, solutions for the assessment of mobile GUIs using machine learning are still unknown. Here we introduce a deep learning model to assess the visual aesthetics of mobile Android applications designed with App Inventor. We used a supervised learning approach to train and compare models using three different architectures. The highest performing model, a Resnet50, achieved a mean squared error of .022. The assessments of new GUIs showed an excellent correlation with human ratings (ρ = .9), and the Bland Altman plot analysis revealed 95% agreement with their labels. These results indicate the model’s effectiveness in automating the visual aesthetics assessment of GUIs of mobile apps.