GUI-based software module clustering through dynamic analysis

Abstract

Software module clustering has been studied to suggest a reference clusters of a complex software system by clustering “cohesive” modules in terms of functionality to facilitate the comprehension and main- tenance of the system. Although various sources of information (e.g., source code) have been used to analyze the cohesiveness in module clustering, there remain several promising sources that have not been studied. A graphical user interface (GUI) is expected to be one such sources since it often contains information of the functionalities of a software system

we generalize the issue to the whole module clustering and introduce a study, called multi-factor module clustering, to cope with the issue.

such information can be useful to find cohesive modules if it is possible to build the links between the GUI and modules. In particular, we utilize the functional information presented on the structure of the displayed GUI and profile the links between the structure and modules by means of dynamic analysis. We demonstrate the feasibility of automated GUI-based module clustering and its merits while applying it to two open-source software systems, one of which is a small- and another is large-scale system. For the small-scale system, we demonstrate its merits in guiding software comprehension based on the manual module clustering by three experts in software engineering. For the large-scale system, we demonstrate it based on the preexisting clusters created by the developers of the system, especially when it compares to other existing approaches. We also derive two extensions of this study and introduce their preliminary studies. In the first study, called GUI-based labeling, we adapt our module clustering approach to introduce an approach providing clues from a GUI for comprehending preexisting module clusters. In the second study, we focus on an significant issue that we found in this study, called single-factor problem