Model-based adaptatıon for end-to-end testing of smart tvs

In this dissertation, we introduce AdapTV, a model-based, black-box, and feedbackdriventest adaptation approach for end-to-end user interface (UI) testing, specificallytargeting smart TVs. Given a test suite known to work on an older version ofthe smart TV and a new version to which the test suite should be adapted, AdapTVbegins by automatically discovering UI models for both the old and new versionsthrough opportunistic crawling. Each test case in the suite is then executed on theolder version, and the corresponding path in the UI model for the old version isidentified. Finally, a semantically equivalent path in the UI model discovered forthe new version of the smart TV is determined and dynamically executed on thenew version in a feedback-driven manner. We empirically evaluate AdapTV in a settingthat closely mimics the setup used in an industrial context, demonstrating thatAdapTV is both effective and efficient, adapting all the test cases with a high successrate. Subsequently, we propose AdapTV+, which enhances AdapTV throughicon recognition. Later, we propose AdapTV 2.0, an AI-centric version of AdapTV,aiming to enhance the original approach with the capabilities of state-of-the-artgenerative AI models. Our evaluations reveal that while generative AI models offerimprovements in certain aspects of AdapTV, they exhibit limitations in others.Also, we introduce GenTV, a model-based test execution approach that translateshigh-level natural language instructions into executable tests. Our evaluation ofGenTV on smart TVs indicates that GenTV effectively and efficiently transformsnatural language instructions into executable tests with a high execution rate.