A Method of Optimizing Multi-Locators Based on Machine Learning

Due to the rapid iteration of Web applications, there are some broken test cases in regression tests. The main reason for the appearance of broken test cases is the failure of element location in the new web page. The element locators in the test cases come from various Web element locating tools, which are used to identify the elements to be convenient for testers to operate them and eventually to test the Web application. Therefore, the Web element locating tools play an essential role in web testing. At present, there are some Web element locating tools, which are supported by a single locating algorithm or multiple locating algorithms. Moreover, the Multi-Locators supported by multiple algorithms are obviously more robust than the one supported by a single algorithm. However, when synthesizing all locating algorithm to generate Multi-Locators, a better method can be selected in assigning weights to each algorithm. Based on this observation, we propose a method to optimize Multi-Locators. In assigning weight to each algorithm, it chooses a weight distribution method based on machine learning, named Learned Weights. Through experimental comparison, it is shown that the locating tool supported by algorithm based on machine learning is more robust than these existing locating tools.