CRAFT: A Framework for Evaluating Software Clustering Results in the Absence of Benchmark Decompositions

Software clustering algorithms are used to create high-level views of a system’s structure using source code-level artifacts. Software clustering is an active area of research that has produced many clustering algorithms. However, we have seen very little work that investigates how the results of these algorithms can be evaluated objectively in the absence of a benchmark decomposition, or without the active participation of the original designers of the system. Ideally, for a given system, an agreed upon reference (benchmark) decomposition of the system’s structure would exist, allowing the results of various clustering algorithms to be compared against it. Since such benchmarks seldom exist, we seek alternative methods to gain confidence in the quality of results produced by software clustering algorithms. In this paper we present a tool that supports the evaluation of software clustering results in the absence of a benchmark decomposition.