Deep features extraction to assess mobile user interfaces

Recently, the quality of user interface design has become a major factor in the success of mobile applications. To this end, the evaluation of Mobile User Interface (MUIs) is a mandatory step. Generally, there exist two main categories of user interface evaluation methods: manual and automatic. The manual evaluation is conducted by users or experts to assess the MUIs and verify whether it works as intended. However, it is more time-consuming task. Automatic evaluation is based on an automatic tool, which needs preconfiguration in the source code. But, this manual configuration is a tedious task for non-programmer evaluators. To address this issue, we propose an evaluation approach based on the analysis of graphical MUI as screenshot without using the source code and user involvement. The proposed approach combines the GoogleNet architecture and K-Nearest Neighbours (KNN) classifier to evaluate the MUIs. First, we apply the Borderline-SMOTE method to obtain a balanced dataset. Then, the GoogleNet is used to extract automatically the features of MUI. Finally, we apply the KNN classifier to classify the MUIs as good or bad. We evaluate this approach based on publicly available large-scale datasets having more number of good ratings than bad ratings which may have an impact on the proposed model during the learning step. To this end, the Borderline-SMOTE method (BSM) is used to obtain a balanced class distribution. The obtained results are very promising.