METHODS: A meta-path-based method for heterogeneous community detection in the open source software ecosystem

Abstract

Detecting communities in the open source software (OSS) ecosystem can help understand the collaborations in the open source software ecosystem and promote an understanding of the dynamics of the ecosystem. However, most existing community detection methods are designed for homogeneous networks, whereas the OSS ecosystem is a heterogeneous network. Therefore, we propose a meta-path-based method for heterogeneous community detection in the OSS ecosystem (METHODS). METHODS comprises four steps. Firstly, a heterogeneous information network is constructed based on meta-paths. Secondly, the Canopy algorithm is used to obtain the number of initial communities. Thirdly, the skip-gram model is used to identify seed nodes for community detection. Finally, METHODS detects heterogeneous communities around the seed nodes. By defining a series of evaluation metrics and verifying these on GitHub datasets, METHODS achieves the best performance of all the other methods. Moreover, the case studies on GitHub also shows METHODS can discover latent communities whose members are potentially relevant.