Finding Critical Users in Social Communities via Graph Convolutions

Finding critical users whose existence keeps a social community cohesive and large is an important issue in social networks. In the literature, such criticalness of a user is measured by the number of followers who will leave the community together when the user leaves. By taking a social community as a k k -core, which can be computed in linear time, the problem of finding critical users is to find a set of nodes, U U , with a user-given size b b in a k k -core community that maximizes the number of nodes (followers) to be deleted from the k k -core when all nodes in U U are deleted. This problem is known to be NP-hard. In the literature, the state-of-the-art approach, a greedy algorithm is proposed with no guarantee on the set of nodes U U found, since there does not exist a submodular function the greedy algorithm can use to get a better answer iteratively. Furthermore, the greedy algorithm designed is to handle k k -core in any social networks such that it does not consider the structural complexity of a given single graph and cannot get the global optimal by the local optimal found in iterations. In this paper, we propose a novel learning-based approach. Distinguished from traditional experience-based