A PageRank based recommender system for identifying key classes in software systems

Program comprehension is a fundamental prerequisite before software engineers may engage in software maintenance or evolution activities and requires the study of large amounts of documentation - either developer documentation or reverse engineered. Very often, from this documentation is missing a short overview document pointing to the most important classes of the system, these who are essential for starting the understanding of the systems architecture. In this work we propose a recommender tool to automatically identify the most important classes of a system. Our approach relies on modeling the static dependencies structure of the system as a graph and applying a graph ranking algorithm. We empirically identify the optimal way of building the system graph, identifying how different dependency types should be taken into account. In experiments performed on a set of open source real life systems, we compare the sets of classes recommended by our tool with these included in the architectural overviews provided by the system developers.