Community detection in software ecosystem by comprehensively evaluating developer cooperation intensity

Abstract

Abstract Context : As soon as the concept of software ecosystem was proposed, it has aroused great interest in both academia and industry. Software ecosystem can be described as a special complex network. Community structures are critical towards understanding not only the network topology but also how the network functions. Traditional community detection algorithms in complex networks mainly utilize the network topology to measure the similarities between nodes. Because of the complexity of information interaction in software ecosystem, only considering the topology structure will lead to unreasonable division of communities. Objective : For solving community detection in software ecosystem more reasonably, we present a method of community detection by comprehensively evaluating developer cooperation intensity in software ecosystems. Method : First, we combine network topology information and developer interaction information to calculate the developer cooperation intensity, so as to deeply explore the relationship between developers from both topological and semantic properties. Then a community detection algorithm ABDCI is proposed based on the cooperation intensity of developers by referring to the hierarchical clustering idea of Louvain algorithm. Finally, this method is applied to many different types of developer networks in the software ecosystem through GitHub hosting platform. Results : Comparing with three classical community detection algorithms, we find that the proposed method can identify a clearer community structure for the developer collaboration network in the software ecosystem. Conclusion : Our approach provides an effective and extensible technique for solving the community detection problem of real developer collaboration network in software ecosystem. According to our findings, we conclude that community detection algorithms based on comprehensive topological properties and semantic properties are more suitable for real communities in software ecosystems than traditional single-property algorithms.