Automating heuristic evaluation of website interfaces using convolutional neural networks

• Main Author: Fernandez, Ryan Austin
• Format: text
• Language: English
• Published: Animo Repository 2019
• Subjects: Deep learning (Machine learning); Heuristic programming; Neural networks (Computer science); Computer Sciences
• Online Access: https://animorepository.dlsu.edu.ph/etd_masteral/7188
• Institution: De La Salle University

Heuristic evaluation is an important phase in both quality assurance and UX de- sign. This ensures that a user interface that is being tested adheres to usabilitystandards and can be used by any user with ease. Heuristic evaluation typicallytakes a long time since it involves consolidating opinions of multiple design ex- perts. This study attempts to automate the detection of usability issues in a givenuser interface design to lessen the expense and thereby time needed to hire pro- fessionals and to focus on development-review-revision cycles. The method usedwas a data-driven approach through the usage of convolutional neural networks (CNN). A computational model using CNNs to determine whether an interface is good or bad is made from a dataset of screenshots of user interfaces, with a higheraccuracy of that of a simple multilayer perceptron. By comparing the model out- puts to evaluator annotations, several insights regarding the design of e-commercewebsites were also gathered, specifically which heuristics are more important tothe models as compared with which are not as important. The highest perform- ing model yielded 70% accuracy. Further research can lead to fine tuning with alarger dataset.