Multi-objective optimization-based overlapping community detection in software ecosystem

Abstract

A software ecosystem can be described as a complex network, consisting of many software projects and stakeholders. In this network, a node may belong to multiple communities, resulting in an overlapping community structure. For a software ecosystem network, overlapping community detection is beneficial to understanding the interaction behaviors between individuals and promoting the healthy development of this system. However, existing methods for overlapping community detection are less accurate or produce less significant community structures. In view of this, we propose a method of overlapping community detection based on multi-objective optimization in a software ecosystem, with domain-specific adaptations. In the proposed method, a multi-objective optimization model for a software ecosystem is first formulated with the maximization of the extended kernel k-means (EKKM) and the extended ratio cut (ERC). Then, a hybrid individual representation that combines character string representation with binary representation based on overlapping communities is developed, and the corresponding crossover and mutation operators are designed to be integrated into the NSGA-II multi-objective optimization framework, which is beneficial to enhancing the population evolution. Six networks in a software ecosystem are built using data collected in GitHub. Based on them, experimental results show the superiority of the proposed method.