Scale Adjustable Interaction Group Identification

Abundant information with rich content is produced by tens of millions of users on social networking services everyday. Users can be clustered different kinds of interaction groups by the topics of their interactions. However, identifying dynamic interaction groups on topics still remains a challenge and the hierarchy of topics is often overlooked. In this paper, we propose a game-theoretic approach based on hierarchical topic model, in order to formulate the dynamics of users' participation into interaction groups formed by users' interrelationships on a social network. Under the assumption that user's partition into interaction groups corresponds to an equilibrium of the game, each user is represented by a selfish agent that chooses to join or exit a group according to its utility which consists a gain function and a loss one. An agent may belong to more than one interaction group because of its several different interests, which is naturally captured by the proposed approach. We also take into consideration the hierarchy of topics, in order to better describe the characteristic of the groups from different levels. The results of experiments which we conduct on Facebook dataset illustrate that the proposed approach is more effective in identifying interaction groups and is able to distinguish these groups on different topic levels and different scales adaptively.