Abstract
This paper proposes the use of Interactive Genetic Algorithms (IGAs) to integrate developer’s knowledge in a re-modularization task. Specifically, the proposed algorithm uses a fitness composed of automatically-evaluated factors—accounting for the modularization quality achieved by the solution—and a human-evaluated factor, penalizing cases where the way re-modularization places components into modules is considered meaningless by the developer.
The proposed approach has been evaluated to re-modularize two software systems, SMOS and GESA. The obtained results indicate that IGA is able to produce solutions that, from a developer’s perspective, are more meaningful than those generated using the full-automated GA. While keeping feedback into account, the approach does not sacrifice the modularization quality, and may work requiring a very limited set of feedback only, thus allowing its application also for large systems without requiring a substantial human effort.