Differences of Structured Documents - Improving their Quality

A classical task of software development environments is the management of document versions. Besides others, primary aims are increasing efficiency and supporting reuse. While for classical textual documents, e.g. source code, versioning is well understood and frequently used, this is not the case for documents with graph structured models as for example the notation forms of the UML. However, these models become more and more important and are like textual documents subject to versioning and co-developed by a number of people. Thus also for graph structured models versioning tools are required. One of the several reasons for the absence of efficient tools is the problem that differences are generally not unique. With the potentially large set of differences evaluable for the same documents, the question of their quality arises. Since for graph structured and in particular for UML documents many different quality measures are reasonable, convincing conclusions can be drawn only by evaluating a set of different measures, contrasting the conventional approaches. This paper presents a novel and more substantive approach of judging differences' qualities and aims at finding optimal differences. Evolutionary algorithms are proposed to solve the resulting multiobjective optimization problems.