| A problem that the software industry frequently faces is the maintenance and improvement of legacy software systems. Though most legacy software systems are still working well, their structure is often no longer understood. When one wants to migrate a legacy software system to a new operating system or different programming language, or to add to its functionality, it is essential to recover the structure of the legacy software system before any change is made on it. Many reverse engineering projects attempt to regain this knowledge.; A common approach to the problem of understanding a large software system is to decompose it into smaller, easier to comprehend subsystems. Though such approaches can aid the process of understanding legacy software systems, an important issue for the current software clustering techniques is that they are hard to evaluate. It is clear that an objective way of comparing different software clustering decompositions is necessary.; In this thesis, we concentrated on comparing different software clustering techniques by comparing their output decompositions. We have improved a method for comparing the output of different software clustering approaches called MoJo and enhanced a metric for evaluating the quality of a software clustering approach. We also introduced a new variation of MoJo that integrates edge information to the MoJo measure.; The approaches presented in this thesis have been implemented and applied to real industrial software systems. The results we obtained demonstrate the effectiveness and usefulness of our techniques. |
