|
Abstract:
|
In the recent researches on network analysis, grouped data has been one of the useful and popular data formats for studying social behavior. Zhao and Weko (2015) have presented an approach called Hub Model to infer the relationships from group data, with the assumption that every observed group has a single leader and the leader has brought together some other members to form a subset of the population. In this study, a generalized model-based approach called Multiple Hubs Model is proposed for grouped data, which assumes that each group can have zero or multiple leaders. This new model allows us to address some questions about the restriction of the applicability due to the one leader assumption in the original Hub Model. A series of simulations based on Hard EM algorithm is applied to test the performance of this new model. We also apply this model into datasets on both animal behavior and human activity.
|
