Our previous approach to test adaptation in smart TVs had a limitation in handling UI elements that lack associated text. To address this issue, we created a dataset and trained a classifier to label frequently used icons found in smart TVs, which can be recognized without an associated text. In cases where a UI element lacks associated text, we utilize the classifier to label the element, enabling the identification of "equivalent states". This paper presents the development, training process, and potential impact of this classifier on our test adaptation technique. Evaluation results show that our classifier significantly improves metrics like success rate, adaptation rate, and test length overhead. Our proposed methodology, combined with the trained classifier, offers practical solutions to enhance test adaptation processes in smart TVs.

In this work, we briefly introduce a model-based test adaptation approach for testing smart TVs produced by Arçelik - the fourth largest home appliances manufacturer in Europe operating in 100 different countries under 10 different brand names, including Beko and Grundig. Although our focus is on smart TVs produced by a single company, the proposed approach can readily be applied to any consumer electronics with a screen-based user interface. This is mainly due to the fact that we present a non-intrusive and completely black-box approach that operates by interpreting the images of user interfaces to interact with the system. More specifically, given a test suite, which is known to work on an older version of the system, and a new version of the system, to which the test cases should be adapted, the proposed approach automatically discovers the user interface models of both the older and the new version of the system by systematically crawling the respective user interfaces; figures out the path traversed by a test case in the model discovered from the old system; dynamically (i.e., in a feedback-driven manner) determines the most "semantically" similar path in the model discovered from the new system; and finally executes the path on the new system. The rationale behind using a model-based approach is to minimize the guesswork (thus, to improve both the effectiveness and the efficiency of the test adaptation) in the presence of significant changes in the user interfaces, such as the ones affecting the order of the screens/interactions.