Competitive and complementary influence maximization in social network: A follower’s perspective

摘要

The problem of influence maximization is to select a small set of seed users in a social network to maximize the spread of influence. Recently, this problem has attracted considerable attention due to its applications in both commercial and social fields, such as product promotion, contagion prevention and public opinion forecasting. Most of prior work focuses on the diffusion model of single propagating entity, purely-complementary entities or purely-competitive entities. However, in reality, the influence diffusion in the social network is certainly more general, involving multiple propagating entities, which are competitive or complementary rather than single entity, purely-complementary entities or purely-competitive entities.In this paper, we consider the problem that a company (follower) intends to promote a new product into the market by maximizing the influence of a social network, where multiple competitive and complementary products have been spreading. We propose a Competitive and Complementary Independent Cascade (CCIC) diffusion model, and propose a novel optimization problem, follower-based influence maximization that aims to select top-K influential nodes as seed nodes, which can maximize the influence of a social network where multiple competitive and complementary products have already been propagated. To solve follower-based influence maximization problem, we propose a Deep Recursive Hybrid model (DRH) and an approximation algorithm (DRHGA). The DRH model dynamically tracks entity correlations, cascade correlations, causalities between ratings and next-period adoption through a deep recursive network and computes influence probabilities between nodes on target product. Then, with the influence probabilities predicted through DRH model, the DRHGA algorithm can efficiently find the seed node set for the target product under the CCIC diffusion model. Experimental results conducted on several public datasets show that our method outperforms the state-of-the-art methods on prediction accuracy and efficiency.