Software Module Clustering Algorithm Research Based Module Dependency Graphic And GEP

Software modularization can improve the understanding of software system and decrease the cost of system Using the software module clustering technique can rebuild the architecture of source codes and obtain the modularized software. Software module clustering technique is based on Module Dependency Graph called MDG. Usually researcher used the MDG of C/C++ source codes. There are few researches about the MDG of Java source codes, so the tools for building MDG of Java source code are scarce. For this reason, the paper presents a tool for building Logic-based Java Model Dependency Graphic, TLJMDG for short. It uses the open source framework JTransformer and open source tool tuProlog to build MDG. It is extendable that researchers can implement the extractor interface to build the MDG extractor for the given MDG and it is also configurable that researchers can set the different weighted value for the different relationship to get the special MDG.To get the good structure of MDG, some researchers use the clustering algorithms to divide the MDGs. The representative approaches are meta-heuristic algorithms, such as Genetic Algorithm and Hill Climbing Algorithm. While these algorithms are easy to converge early and cann’t get the global optional solution. For this reason, this paper presents an improved GEP algorithm called GEP-SC algorithm for the software module clustering. This algorithm uses the traditional GEP framework, gives the limitation for the traditional crossover operator and mutation operator in order to guarantee the integrity of MDG and nodes. Experimental results show GEP-SC outperforms hill climbing.