Applying Social Network Analysis Techniques to Architectural Smell Prediction

As a software system evolves, the amount and complexity of the interactions amongst its components is likely to increase, which negatively affects the system design structure and also its quality. For instance, certain modules might become coupled due to a new user feature being added or to suboptimal development decisions. Design degradation symptoms are often related to high coupling and unwanted dependencies, such as: cyclic dependencies or violations to design rules, amongst other architectural smells. Thus, the early detection of such symptoms is important for architects to: i) anticipate dependency-related design problems in different parts of the system, ii) assess possible situations of technical debt, and iii) proactively look for solutions to preserve the quality of the system. Although there are approaches that analyse design dependencies in code bases and flag smell occurrences, very few of them have dealt with the prediction of dependency relations amongst software components. This research hypothesises that a predictive approach can warn architects about dependency-related problems before they appear. To this end, a particular graph-based approach is social networks analysis (SNA), which has been used for modelling both nature and human phenomena. Specifically, SNA techniques can predict links that do not yet exist between pairs of nodes in a network. SNA applications have shown evidence that the topological features of dependency graphs can reveal interesting properties of the software system under analysis. Nonetheless, SNA techniques have not yet been extensively exploited in the Software Architecture community. In this context, the question that motivates this research is to what extent SNA can leverage on information from a software design (and its evolution over time) for inferring new dependencies and likely configurations of architectural smells out of those dependencies.