Automatic identification of class stereotypes

Abstract

An approach is presented to automatically determine a class's stereotype. The stereotype is based on the frequency and distribution of method stereotypes in the class. Method stereotypes are automatically determined using a defined taxonomy given in previous work. The stereotypes, boundary, control and entity are used as a basis but refined based on an empirical investigation of 21 systems. A number of heuristics, derived from empirical evidence, are used to determine a class's stereotype. For example, the prominence of certain types of methods can indicate a class's main role. The approach is applied to five open source systems and evaluated. The results show that 95% of the classes are stereotyped by the approach. Additionally, developers (via manual inspection) agreed with the approach's results.