Mining instances of structural design patterns from class diagrams based on sub-patterns

Abstract

In order to improve the quality of a software system and reuse expert experience in software system design, design patterns have been extensively applied in the software industry. Mining design pattern instances from source codes can assist the understanding of the systems. In this paper, we propose eight sub-patterns based on common structural features of structural design patterns and their variants. We introduce two kinds of graph, of which one represents the system design and the other describes the sub-patterns. Our approach first selects the candidate classes in UML class diagrams to form the sub-graphs of the system graph, which are then checked isomorphic to the sub-pattern graph or not. These isomorphic sub-graphs are regarded as corresponding to instances of the sub-patterns. After that, we combine some relevant sub-patterns and compare the collectives with structural feature models. For these matched ones, we then apply behavioral analysis to obtain the final pattern instances. The results of the experiments demonstrate that our approach obtains better precision than the existing approaches. © 2013 Springer-Verlag Berlin Heidelberg.