Moderate Detection and Removal of Omnipresent Modules in Software Clustering

Software clustering is important in program comprehension, and it is applied to architecture recovery and software visualization to analyze the actual software system structure. The detection of omnipresent modules, programs used by many parts of the system such as utility classes, is a well-known issue in software clustering. Omnipresent modules obscure the system structure. The detection and removal of omnipresent modules have been studied for years. However, traditional approaches based on the number of the caller classes remove too many classes, including important domain model classes. In this study, we introduce a method to detect omnipresent modules by calculating a metric called usage equality; it represents the equality in the usage of a class by its caller classes based on the dedication score, which is a weight calculated for inter-class dependency defined in the literature. The developed method was evaluated and compared with traditional methods using real examples. The results show that our method identified omnipresent modules more moderately than the traditional methods to avoid the essential domain classes.