Similarity assessment of UML class diagrams using simulated annealing

As model driven development has been promoted to the focus of engineers during the software development, engineers find themselves dealing with a large collection of models. Without managing these models efficiently, the wheel is reinvented over and over, resulting in more duplicated artifacts and an aggravated maintenance effort. Models´ matching is a basic operation for different model management operations such as models´ consolidation and retrieval. It is a kind of a combinatorial problem. The difficulty of the problem comes in two main streams, the similarity assessment and the matching complexity. In this paper, we present the use of Simulated Annealing for matching UML class diagrams based on their lexical, internal, neighborhood similarity, and a combination of them. Additionally the paper empirically compares the accuracy of different similarity metrics in capturing the similarity between two class diagrams within and across domains.