Concept analysis techniques for software modularization

Abstract

Modularity is one of the most important principles in software engineering. According to modularity principle, the structure of software is decomposed into a set of relatively independent modules that hide inner details and only expose relevant information. The information hiding of modules supports to manage the complexity of software development. Among various proposals for automatic modularization of software designs, we follow the line of concept-based approaches that use the theory of Formal Concept Analysis (FCA). These approaches provide an intent, in the form of propositional formula, of an identified module and this intent is useful to interpret the meaning of the module. However, to apply concept-based approaches in large-scale software designs, the \emph{granularity problem} should be addressed. Roughly speaking, FCA generally results in a relatively fine granularity of modules because of the conjunctive characterization of formal concepts and, thus, may be inadequate for large-scale designs. To address this problem, we investigate recent proposals for concept formulations, named Property-Oriented Concept Analysis (POCA) and Object-Oriented Concept Analysis (OOCA). This investigation reveals that these recent concept formulations can provide \emph{coarser} modules than those of FCA. In addition, the formulation of OOCA guarantees the encapsulation of intents in identified modules and, thus, OOCA is superior to other concept formulations with respect to the principle of information hiding. Consequently, we decide the employment of OOCA in software modularization. For practicality, we employ Genetic Algorithm (GA) and design a chromosome representation to avoid the explicit construction of a concept lattice and the enumeration of concept partitions. Because there are generally very large number of concept partitions, it is inevitable to adopt a search technique for finding plausible concept partitions within a reasonable time. In ad...