Commit 30b73305 authored by Aldo Von Wangenheim's avatar Aldo Von Wangenheim
Browse files


Visual aesthetics is seen as an essential factor of usability, interaction, and the appraisal of user interfaces, especially with respect to mobile applications.
Recently, machine learning approaches have shown great promise in being able to predict visual aesthetics. First solutions for the automatic assessment of visual aesthetics are focusing on web user interfaces only. Therefore, we present a deep learning approach to automatically quantify the visual aesthetics of Android mobile user interfaces (MUIs) adopting a regression-based approach.
We employ pairs of Android app screen images and user rating to train a convolutional neural network (CNN) adopting a supervised machine learning approach. For the testing procedure, the ground truth of each Android app screen is the mean score of all user ratings it has received. Our evaluation results demonstrate that a convolutional neural network can learn the prediction of visual aesthetics of MUIs based on the images of the screenshots with a mean squared error of 0.051369 with the test set of 630 screenshots.
The predictions from the Appsthetics model are highly correlated with human ratings (pearson correlation coefficient r = 0.74, p < 2.2e-16) and the Bland & Altman analysis indicate that more than 96% of them agree.
These promising results indicate that our method can serve as an effective and efficient means for providing objective aesthetics evaluation during the design process.
Keywords: Aesthetics, Mobile application, Android, Deep learning, Visual design
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment