Feature Detection in Ajax-enabled Web Applications

In this paper we propose a method for reverse engineering the features of Ajax-enabled web applications. The method first collects instances of the DOM trees underlying the application web pages, using a state-of-the-art crawling framework. Then, it clusters these instances into groups, corresponding to distinct features of the application. The contribution of this paper lies in the novel DOM-tree similarity metric of the clustering step, which makes a distinction between simple and composite structural changes. We have evaluated our method on three real web applications. In all three cases, the proposed distance metric leads to a number of clusters that is closer to the actual number of features and classifies web page instances into these feature-specific clusters more accurately than other traditional distance metrics. We therefore conclude that it is a reliable distance metric for reverse engineering the features of Ajax-enabled web applications.