文摘
Complexity networks, such as social networks, biological networks and co-citation networks, are ubiquitous in reality. Identifying groups from data is critical for network analysis, for it can offer deep insights in understanding the structural properties and functions of complex networks. Over the past decades, many endeavors from interdisciplinary fields have been attempted to identify groups from data. However, little attention has been paid on exploring groups and their relationships from different views. In this work, we address this issue by using canonical correlation analysis (CCA) to analyze groups and their interplays in the networks. To further improve the interpretability of results, we solve the optimization problem with sparse learning, and then propose a generalized framework of group discovery from heterogeneous data. This framework enables us to find groups and explicitly model their relationships from diverse views simultaneously. Extensive experimental studies conducted on both synthetic and DBLP datasets demonstrate the effectiveness of the proposed method.