Using Library Dependencies for Clustering

Abstract

Software clustering is an established approach to automatic architecture recovery. It groups components that are in some way similar to each other. Usually, the similarity measure is based on the dependencies between components of a system. Library dependencies are intentionally ignored during the clustering process – otherwise, system components would be clustered with library components they use. We propose to particularly look at the dependencies on external components or libraries to learn more about an application’s high-level structure. The number of dependencies of a component from different kinds of libraries provides insightful information about the component’s functionality. Our case study illustrates the potential of this idea.