Multi-stage opinion maximization in social networks

Opinion maximization is a crucial optimization approach, which can be used in preventative health, such as heart disease, stroke or diabetes. The key issue of opinion maximization is to select a subset of initial influential individuals (i.e., seed nodes) with the desired opinion, spread the desired opinion to their neighbors and achieve the greatest opinion propagation. Previous researches on opinion maximization focus on user's activation process or static opinions, but pay little attention to the combination between their activation process and dynamic opinion process. In addition, prevalent greedy and heuristic algorithms have some disadvantages, such as low accuracy or low efficiency. In this paper, we study the multi-stage opinion maximization for preventative health in social networks. First, we formulate the opinion maximization problem and leverage the proportion of desired opinions as the objective function. Based on multi-stage independent cascade and weighted voter model, we design the activated voter model to obtain user's activation status and dynamic opinion process. Moreover, we propose a novel Multi-stage Opinion Maximization Scheme (MOMS), which is composed of three phases: (i) the selection of candidate seed nodes, (ii) the generation of seed nodes and (iii) dynamic change of node opinions by the activated voter model. We use an effective heuristic rule to exclude some less essential nodes and select candidate seed nodes. Then, we determine seed nodes of each stage using the improved heuristic algorithm through combining the advantages of heuristic algorithm and greedy algorithm. Finally, experimental results on six social network datasets demonstrate that the proposed method has more superior proportion of desired opinions than the chosen benchmarks.