Revealing the relationship between architectural elements and source code characteristics

Understanding how a software system is structured, i.e. its architecture, is crucial for software comprehension. It allows developers to understand an implemented system and reason about how non-functional requirements are addressed. Yet, many systems lack any architectural documentation, or it is often outdated due to software evolution. In current practice, the process of recovering a system's architecture relies primarily on developer knowledge. Although existing architecture recovery approaches can help to identify architectural elements, these approaches require improvement to identify architectural concepts of a system automatically. Towards this goal, we analyze the usefulness of adopting different code-level characteristics to group elements into architectural modules. Our main contributions are an evaluation of the relationships between different sets of characteristics and their corresponding accuracies, and the evaluation results, which help us to understand which characteristics reveal information about the source code structure. Our experiment shows that an identified set of characteristics achieves an average accuracy of 80%, which indicates the usefulness of the considered characteristics for architecture recovery and thus to improving software comprehension.