An Improved Group-based Influence Maximization Method in Social Networks

Influence maximization is a classic problem studied in social network analysis. This problem is NP-hard and can be solved with a greedy algorithm. However, the method requires tens of thousands of Monte-Carlo simulations, which are very time consuming and not scalable. To improve this method, researchers presented the community-based influence maximization method. However, this method detects communities based on node connections while generally ignores the influence property of nodes. In addition, when computing the influence spread based on community structure, it loses sight of the community size and border nodes. To improve the community-based influence maximization method, this paper first finds groups with similar influence characteristics based on the influence property of nodes. Then influence spread is approximately calculated based on the group structure in which the group size and border nodes are considered. Experiments demonstrate that the group-based influence maximization method in this paper achieves better influence spread than corresponding community-based influence maximization methods with matching running time.