KuiTest: Leveraging Knowledge in the Wild as GUI Testing Oracle for Mobile Apps

In industrial practice, UI (User Interface) functional bugs typically manifest as inconsistent UI input and corresponding response. Such bugs can deteriorate user experiences and are, therefore, a major target of industrial testing practice. For a long time, testing for UI functional bugs has relied on rule-based methods, which are labor-intensive for rule development and maintenance. Given that the UI functional bugs typically manifest where an app's response deviates from the user's expectations, we proposed the key point of reducing human efforts lies in simulating human expectations. Due to the vast in-the-wild knowledge of large language models (LLMs), they are well-suited for this simulation. By leveraging LLMs as UI testing oracle, we proposed KuiTest, the first rule-free UI functional testing tool we designed for Meituan, one of the largest E-commerce app providers serving over 600 million users. KuiTest can automatically predict the effect of UI inputs and verify the post-interaction UI response. We evaluate the design of KuiTest via a set of ablation experiments. Moreover, real-world deployments demonstrate that KuiTest can effectively detect previously unknown UI functional bugs and significantly improve the efficiency of GUI testing.