A Conditional Entropy-Based Evaluation of Clustering Approaches to Automatic Web Services Refactoring

Service-Oriented Architecture has been widely adopted in software development because of its properties of addressing distributed and heterogeneous environments. At the same time, the highly-demanding flexibility and capacity of adaptation in software projects introduce a process of constant code modifications and complexity growth. Thus, developers quickly make mistakes such as code duplication or unnecessary code, negatively impacting quality attributes such as performance and maintainability. Those modifications are usually conducted by refactoring, which significantly improves software quality. In this context, we aim to compare manual service groupings and automatic groupings that allow analyzing, evaluating, and validating clustering techniques applied to improve service cohesion and fragmentation in SOA-driven software projects. Concerning quantifying the quality of our clustering solutions, we relied on external quality metrics, which measure how well the clustering results match some prior knowledge (manual service groupings by real developers in our case) about the data. Additionally, we have performed improvements in existing clustering techniques of previous work, VizSOC, reaching a 20% gain regarding the metrics mentioned above. Moreover, we added an implementation of the COBWEB clustering algorithm to expand our research, yielding fruitful results.