Um Clustering na Recuperacao de Componentes de Software

Abstract

Actually, there are Software Systems without documentation, especially when they are legacy systems. Exists many proposed methods to get a structure abstraction from these systems. The methods found in the literature to do this work are based in clustering, because the similar objectives between a software system structure and a clustering process: the software modules must to be high cohesive and low coupling, in similar way a cluster must to have similar objects and different from other clusters. The methods are based mainly in hierarchical clustering. In this work we propose an adaptation of KMeans algorithm inside the GRASP framework known as the GraspKM algorithm, to obtain an abstraction from software systems. This method solves the clustering as a combinatorial( optimization problem, and it is efficient optimizing the objective function proposed.