ITeM: An Intent-Driven Migration Framework for Mobile App GUI Testing

ITeM is an innovative intent-driven GUI test migration method that uses the understanding and reasoning abilities of large language models. It employs a two-stage framework of intent generation and dynamic reasoning to overcome the limitations of traditional control-matching-based test migration methods when interaction logic changes. This method has been experimentally validated for its effectiveness and efficiency advantages and was published at ISSTA 2025, a top conference in the software engineering field.

Mobile app GUI testing is a core means to ensure quality, but manual test case construction is costly and labor-intensive, driving the development of automated test migration technologies. Traditional methods model test migration as a control matching task, which performs well when interaction logic is consistent but fails when interaction logic changes between different apps—yet interaction logic variations are extremely common in the actual mobile app ecosystem.

ITeM uses a two-stage framework:

1. Intent Generation: Analyze the test execution trajectory of the source app through a transition-aware mechanism to abstract the tester's real task intent, decoupling test logic from UI implementation;
2. Dynamic Reasoning Execution: Based on the current interface state of the target app, use large language models to infer the specific operational steps needed to fulfill the intent, naturally adapting to changes in interaction logic.

Tech Stack: Appium 2.2.1 (automation control), Android SDK and emulators (supports 6.0/11.0), Java8/Python3.11 (core languages), OpenAI API (LLM integration). System Modules:

• Execution Tracking: Records the test execution trajectory of the source app;
• Intent Generation: Generates high-level intents based on the trajectory;
• Intent Migration: Implements the intent in the target app;
• Assertion Migration: Migrates test assertions and verification logic.

The research team conducted experiments on 35 real Android apps with 280 test migration tasks. The results show that ITeM has significant advantages in both effectiveness and efficiency compared to the current state-of-the-art methods. Its core robustness comes from migration at the semantic level (intent) rather than the syntactic level (control), fundamentally avoiding the matching failure problem caused by changes in interaction logic.

The ITeM research成果 was published at ISSTA 2025, a top conference in software engineering. The paper details the method design, implementation details, and experimental evaluation. The relevant dataset has been open-sourced on the Zenodo platform to facilitate reproduction and extension by other researchers.

For mobile app development teams, ITeM can significantly reduce test maintenance costs during UI refactoring/function adjustments and improve the reusability of test assets. From a macro perspective, ITeM demonstrates the deep application potential of large language models in complex software engineering tasks (such as test automation), pointing the way for the development of intelligent testing tools.