Studying the evolution of software through software clustering and concept analysis

Abstract

This thesis describes an investigation into the use of software clustering and concept analysis techniques for studying the evolution of software. These techniques produce representations of software systems by clustering similar entities in the system together. The software engineering community has used these techniques for a number of different reasons but this is the first study to investigate their uses for evolution. The representations produced by software clustering and concept analysis techniques can be used to trace changes to a software system over a number of different versions of the system. This information can be used by system maintainers to identify worrying evolutionary trends or assess a proposed change by comparing it to the effects of an earlier, similar change. The work described here attempts to establish whether the use of software clustering and concept analysis techniques for studying the evolution of software is worth pursuing. Four techniques, chosen based on an extensive literature survey of the field, have been used to create representations of versions of a test software system. These representations have been examined to assess whether any observations about the evolution of the system can be drawn from them. The results are positive and it is thought that evolution of software systems could be studied by using these techniques.