Research And Implementation Of Influence Maximization And Parallelization Across Multiple Networks

Online Social Networks(OSNs)have become an imperative channel for people to obtain and publish information.Variety of social networks provide a range of services.Plenty of individuals are getting involved in more than one social networks,and maintaining multiple relationships of social networks.The value behind the integrated information of multiple social networks is high.Therefore,it's necessary to identify a small subset of seed nodes who can make the influence maximized based on multiple social networks.However,pervious works could not improve the efficiency and accuracy at the same time.MapReduce and BSP which are popular framework for parallel-distributed computing,are designed for large scale neworks analysis and processing.BSP is fit for iterative graph algorithm with obvious heterosis.The main work of this paper is as follows:1.The compule of multiple social networks in the previous research is not reasonable with the assumption of shared users will propagate all the information reaching them to the other networks.In this paper,we take the shared users' s complicated behaviors into account and propose a new compule for multiple social networks.2.The previous approaches are not efficent and scalable enough for large scale socail networks.This paper tries to improve the scalability and performance of greedy algorithm on influence maximization,and propose an inflence maximazition algorithm based on BSP(BSPGreedy).3.One of important things in parallel-distributed computing for large scale social networks is graph partitioning.For the purpose of parallel-distributed computing,this paper propose a multilevel framework for multiple graph partitioning(CPMN).