Clustering Similarity Measures for Architecture Recovery of Evolving Software
Automated software architecture recovery of module views from source code is a challenging research issue. Different similarity measures are used to evaluate clustering algorithms in the software architecture recovery of module views. However, few studies seek to evaluate whether such measures accurately capture the similarities between two clusterings. This work presents an evaluation of six clustering similarity measures through the use of intrinsic quality and stability measures and the use of ground truth architectures proposed by developers. The results suggest that the MeCl metric is the most adequate to measure similarity in the context of comparison with ground truth models provided by developers. However, when the architectural models do not exist, the Purity metric shows the best results, as measured by the correlation with the intrinsic Silhouette coefficient.
Garcia, J., Ivkovic, I., and Medvidovic, N. (2013). A Comparative Analysis of Software Architecture Recovery Techniques. In Int’l Conf. Automated Software Engineering.
Lutellier, T., Chollak, D., Garcia, J., Rayside, D., Kroeger, R., Tan, L., Rayside, D., Medvidovic, N., and Kroeger, R. (2015). Comparing Software Architecture Recovery Techniques Using Accurate Dependencies. In 37th Int’l Conf. Software Engineering.
Mitchell, B. S. and Mancoridis, S. (2001). Comparing the decompositions produced by software clustering algorithms using similarity measurements. In International Conference on Software Maintenance.
Tzerpos, V. and Holt, R. C. (1999). MoJo: a distance metric for software clusterings. In 6th Working Conference on Reverse Engineering.
Wu, J., Hassan, A. E., and Holt, R. C. (2005). Comparison of Clustering Algorithms in the Context of Software Evolution. In 21st International Conf. on Software Maintenance.