A Novel Unsupervised Learning Approach for Assessing Web Services Refactoring

During the last years, the development of Service-Oriented applications has become a trend. Given the characteristics and challenges posed by current systems, it has become essential to adopt this solution since it provides a great performance in distributed and heterogeneous environments. At the same time, the necessity of flexibility and great capacity of adaptation introduce a process of constant modifications and growth. Thus, developers easily make mistakes such as code duplication or unnecessary code, generating a negative impact on quality attributes such as performance and maintainability. Refactoring is considered a technique that greatly improves the quality of software and provides a solution to this issue. In this context, our work proposes an approach for comparing manual service groupings and automatic groupings that allows analyzing, evaluating and validating clustering techniques applied to improve service cohesion and fragmentation. We used V-Measure with homogeneity and completeness as the evaluation metrics. Additionally, we have performed improvements in existing clustering techniques of a previous work, VizSOC, that reach 20% of gain regarding the aforementioned metrics. Moreover, we added an implementation of the COBWEB clustering algorithm yielding fruitful results.