Evaluation of dependency blossoms on the Maven Central Repository

Abstract

Reusing code is a common practice by leveraging dependencies, libraries, or frameworks. Yet, there is little to no research targeting framework detection on the Maven Central Repository. One proposed way to detect those frameworks are dependency blossoms consisting of artifacts sharing a common group and the same version. Our goal is to evaluate this definition by providing our own, less strict definition of dependency blossoms. Our approach is based on the assumption that artifacts belonging to the same framework will frequently be published within a small and shared time window. We further propose a pipeline on how to compile the blossoms and implemented parts of the pipeline. In order to test our implementation, we ran it on 10,000 randomly selected releases provided by the Goblin framework dataset. We find that 75% of dependency blossoms, using our approach, consist of a minimum of 1 and up to 6 artifact groups for certainty thresholds of 0.8 or higher. This certainty threshold indicates how certain we are that the given artifact is part of the blossom. Thus, our results indicate that the restriction of dependency blossoms to only consist of artifacts that share a common group and the same version is too strict.