LAUSR

AbstractSoftware modularization techniques are employed to understand a software system. The purpose of modularization is to decompose a software system from a source code into meaningful and understandable subsystems (modules). Since modularization of a software system is an NP-hard problem, the modularization quality obtained using evolutionary algorithms is more reasonable than greedy algorithms. All evolutionary algorithms presented for software modularization only consider structural features that are dependent on the syntax of programming languages. For most programming languages, does not exist a tool to extract structural features, so it is not possible to modularize them. To overcome this problem, this paper presents a new multi-objective fitness function, named MOF, which exploits the structural (such as calling dependency and inheritance dependency) and non-structural features (such as semantic contained in the code comments and identifier names), aiming to automatically guide optimization algorithms to find a good decomposition of software systems. To evaluate the performance of this objective function, three optimization strategies, namely global-based search, combining global and local search, and Estimation of Distribution (EoD), are adapted to optimize it. The results on Mozilla Firefox indicate that the proposed algorithm which is based on EoD along with the new MOF function outperforms the algorithms that use structural-based objective functions in guiding the optimization process, in finding more understandable modules.