Imitation Learning Based GUI Automation Testing for Android Application

To address the issue that current reinforcement learning-based Android application GUI automation testing methods fail to comprehend the application's business logic, resulting in suboptimal testing efficiency, this paper proposes an automated GUI testing method for Android applications based on imitation learning. The expert behavior traces are first formalized and collected, followed by the application of the PrefixSpan algorithm to extract behavioral patterns. The extracted patterns are subsequently used to initialize Q-values, which are then integrated with random exploration and Upper Confidence Bound (UCB) based event selection strategies to generate test inputs, prioritizing the exploration of application-related functions. Experimental results indicate that the proposed method demonstrates improvements in test coverage and efficiency when compared to existing Android application GUI testing methods.