Leveraging keyword-guided exploration to build test models for web applications

Abstract Context Dynamic exploration techniques, which automatically exercise possible user interface elements, have been used to explore user interface state flow graphs as test models for web applications. An exhaustive exploration may incur the well-known state explosion problem. In a limited amount of time, most existing dynamic exploration techniques tend to become mired in local or irrelevant regions of the web application due to not considering functionality semantics information. Hence, generated test models have often inadequate functionality coverage for deriving effective test cases. Objective This paper proposes a keyword-guided exploration strategy for automatic construction of web application test models. The goal is to generate incomplete test models with adequate functionality coverage in a given time budget for deriving test cases w.r.t. specified functionalities. Method Given very few keywords that describe specified functionalities, our strategy guides the exploration to discover user interface states and transitions among them that are relevant to the specified functionalities by computing similarity scores between text contents in web pages and given keywords. We use nine representative web applications to perform dynamic explorations in a given time budget and empirically evaluate functionality coverage, and other metrics, e.g., code coverage, the size of test model, the number of the test suite, path diversity, and DOM diversity. Results Our keyword-guided exploration strategy achieves a higher functionality coverage as compared with the generic and feedback-directed exploration strategies. Yet the significant improvement of functionality coverage achieved by our strategy is not exchanged at the cost of other metrics. Conclusion Our keyword-guided exploration strategy is more effective than the generic and feedback-directed exploration strategies in terms of functionality coverage. In a limited amount of time, test models generated with our strategy can be used to derive effective web application test cases.