Learning Effective Oracle Comparator Combinations for Web Applications

Web application testers need automated, effective approaches to validate the test results of complex, evolving Web applications. In previous work, we developed a suite of automated oracle comparators that focus on specific characteristics of a Web application's HTML response. We found that oracle comparators' effectiveness depends on the application's behavior. We also found that by combining the results of two oracle comparators, we could achieve better effectiveness than using a single oracle comparator alone. However, selecting the most effective oracle combination from the large suite of comparators is difficult. In this paper, we propose applying decision tree learning to identify the best combination of oracle comparators, based on the tester's effectiveness goals. Using decision tree learning, we train separately on four Web applications and identify the most effective oracle comparator for each application. We evaluate the learned comparators' effectiveness in a case study and propose a process for testers to apply our learning approach in practice.