On The Significance of Relationship Directions in Clustering Algorithms for Reverse Engineering

Abstract: Software clustering is a common technique applied to simplify reverse engineered software models. These algorithms commonly classify similarity between nodes based on their relationships. However little research exists that discusses the importance of the direction of these relationships. In this paper we provide empirical data for how treating direction in entity relationships affect the recovery accuracy of hierarchical clustering algorithms. We test variations of a hierarchical clustering algorithm on several open-source systems and compare their results, and conclude that relationship direction does not have an significant impact on recovery accuracy. As such, researchers may opt to implement hierarchical clustering algorithms using only one direction of relations instead of both, and still get similar results for less computational cost.