Text2test: From Natural Language Descriptions To Executable Test Cases Using Named Entity Recognition

Official URL: https://risc01.sabanciuniv.edu/record=b3205814

In this work, we present text2test, an innovative approach for automated testing of mobile application user interfaces (UIs). As mobile applications become increasingly prevalent, ensuring robust and user-friendly UIs has become essential, leading to a greater need for efficient testing methodologies. However, testing UIs poses challenges due to varying screen sizes, evolving UI elements across versions, and the need for frequent test case revisions. To address these challenges, we propose text2test, which combines named entity recognition (NER) and semantic similarity computations in a framework using Android APIs to execute test cases from natural language descriptions. We bridge the gap between textual input and UI interactions by training a NER language model to identify UI elements and actions from natural language test case descriptions. Using the DOM structure of an application, containing XML metadata of UI elements, we accurately detect the appropriate UI element associated with the action. Finally, using the information extracted by the NER model and the elements detected using semantic similarity we developed a framework that can execute test cases on Android applications. Our experiments show that text2test achieves a 92% precision rate in identifying element-action pairs and an average accuracy of 88% in detecting expected UI elements and a 76% success rate to fully reproduce test cases on Android applications. Our approach streamlines automated UI testing, reducing manual intervention and the need for frequent script updates and promises a solution for efficient and reliable UI testing.