Matching UML class models using graph edit distance

Abstract

The Unified Modelling Language (UML) class model is an essential constituent in the software system development process and a considerable body of knowledge is encompassed in the form of class model designs. A UML class model forms an elaborate specification hierarchy and comparing different class models in order to identify corresponding parts assumes considerable human expertise. To imitate such human capacity an exponentially complex task needs to be addressed. Yet, the research that involves UML class model matching focuses primarily only on a design pattern detection and studies that tackle the problem of matching any class models are rather rare. The aim of this study is to introduce a class model distance computation framework that can be utilised for comparing class models in model repositories. The framework exploits the relational structure between model elements as well as internal element features to devise a distance measure between any pair of class models. The relational structures of two class models in the form of graphs are aligned using the graph edit distance technique. The internal element feature distance computation deploys the Hungarian algorithm for optimal assignment of any two-feature sets. The distance computation framework reduces the comparison task to polynomial time complexity. The study presents experimental performance analysis of the proposed framework conducted using the precision-recall and receiver operating characteristics curves and corresponding areas under the curves. The results of the analysis indicate low false positive rates for both pairwise and pattern detection tasks.