Group Identity Matching Across Heterogeneous Social Networks

User identity linkage aims to identify and link users across different heterogeneous social networks. In real applications, one person's attributes and behaviors in different platforms are not always same so it's hard to link users using the existing algorithms. In this paper, we discuss a novel problem, namely Group Identity Matching, which identifies and links users by an unit of group. We propose an efficient approach to this problem and it can take both users' behaviors and relationships into consideration. The algorithm incorporates three components. The first part is behavior learning, which models the group's behavior distribution. The second part is behavior transfer and it optimizes the behavior distance between groups across the social networks. The third part is relationship transfer and it enhances the similarity of the groups' social network structure. We find an efficient way to optimize the objective function and it convergences fast. Extensive experiments on real datasets manifest that our proposed approach outperforms the comparable algorithms.