Group-Based Competitive Influence Maximization

Abstract

Traditional competitive influence maximization (CIM) problem in social networks usually considers all small communities as a single global community and calculates the number of influenced nodes to represent the influence spread, which neglects the effect of the groups/communities. In order to solve the loss of local effects in CIM problem, we introduce the concept of the group and divide the local effects into two types, namely the intra-group effect and the inter-group effect. We identify this kind of problem as the group-based competitive influence maximization(GCIM) problem and design an effective method to estimate the local effects. We prove that the GCIM problem under the Competitive Group-Linear Threshold (CG-LT) model is NP-hard. To ensure the reasonableness of data set partition, we make the size distribution of the processed communities be basically the same as the real-world. Experimental results show that our algorithm performs well in real-world social networks of different scales. Moreover, we find that the local effect is closely related to the speed of communication in the network.