Mobile User Interface (UI) serves as a major window where the communication between users and mobile applications happens. It not only defines the look and feel of an app but also plays a key role in creating good interactive experience with the installed functions and contents for users. While designers strike to craft a good UI, a potential gap between designers’ intention and users’ perceived quality of the design might appear. Therefore, understanding how users perceive the UI design, e.g., the perceived usability and aesthetics, is crucial for designers to reflect on and reshape their products for better user experience. It requires designers to frequently elicit feedback from target users and/or domain experts during the iterative app design process. Although deemed effect...[ Read more ]

Mobile User Interface (UI) serves as a major window where the communication between users and mobile applications happens. It not only defines the look and feel of an app but also plays a key role in creating good interactive experience with the installed functions and contents for users. While designers strike to craft a good UI, a potential gap between designers’ intention and users’ perceived quality of the design might appear. Therefore, understanding how users perceive the UI design, e.g., the perceived usability and aesthetics, is crucial for designers to reflect on and reshape their products for better user experience. It requires designers to frequently elicit feedback from target users and/or domain experts during the iterative app design process. Although deemed effective, this approach is resource-intensive. In contrast, this thesis explores the use of data-driven methods to model user perception towards mobile UI design and further support the generation of more usable UI. We first propose a prediction model to infer the perceived brand personality of mobile apps from their static UI pages. In particular, we compile a set of color-based, texture-based, and organization-based visual descriptors of UI pages and demonstrate their promising predictive power with a non-linear prediction model on a collected dataset. The results can benefit designers by highlighting contributing graphical factors to brand personality creation. Next, to analyze the dynamic UI changes, i.e., mobile UI animation, we introduce a two-stream deep neural network to model the user engagement with UI animation, which shows a reasonable accuracy. Based on the features encoded by the model, we further derive the potential design issues of animation to inform design improvement. We develop a prototype AniLens and evaluate it with professional designers. Finally, we investigate how computational powers can aid designers in generating more user-friendly mobile UI. We leverage online curation data to generate the perceived semantics of color filters. Our results indicate that the mobile UI of color filter applications incorporated with the derived semantics which is in line with users’ consensus, can achieve better user experience. In all, we demonstrate the reasonable effectiveness of our proposed data-driven methods in modeling user perception towards mobile UI and also provide insight into how they can be leveraged to facilitate UI generation. At the end, we conclude the thesis by sketching the future work on developing more supportive computational tools for mobile UI design.