“What You See Is What You Test”: Recommending Features from GUIs for Requirements-Based Testing

Abstract

Requirements-based testing (RBT) advocates the design of test cases in order to adequately exercise the behavior of a software system without regard to the internal details of the implementation. To address the challenge that requirements descriptions may be inaccurate in practice, we align requirements engineering and software testing in a novel way by not counting on a complete and up-to-date requirements documentation. Rather, we maintain the black-box nature of RBT to recommend features as the units of testing from software’s graphical user interfaces (GUIs). In particular, we exploit optical character recognition (OCR) to identify the textual information from GUIs, and further build the GUI-feature correspondences based on software’s user-centric documentation which may exhibit partial correctness. Such correspondences from multiple software systems in the same domain serve as a foundation for our recommendation engine, which suggests the to-be-tested features related to a given GUI. We demonstrate our recommender’s feasibility with a study of five products in the web conferencing domain, and the results show the more complete set of features against which a GUI needs to be tested.