Research On Influence Maximization Propagation Models And Its Algorithms In Social Network

The rapid development of social networks has profoundly changed the way people live in today's society.People are getting used to sharing details of their lives on social networks and are more willing to obtain information of interest from social networks.How to use the user group of social network to carry out data mining and commercial marketing has also become the focus of current research.This thesis conducts research on influence maximization in the context of viral marketing.The research content can be divided into two parts,the membership based influence maximization in social network and the problem of potential value based influence maximization with limited cost.The main contents are as follows:(1)The membership based influence maximization in social networkThis thesis introduces the membership into the study of influence maximization.By analyzing the difference between membership marketing and traditional physical marketing,this thesis divides the members into ordinary members and sticky members,and clarifies the goal of membership-based marketing in pursuit of maximizing the number of sticky members.Based on this,an influence propagation called MBIC(Membership Based Influence Cascade)which models the process of ordinary users accepting friend recommendations and eventually becoming sticky members is proposed.MBIC introduces the analysis of the activity and intimacy of social network users and divides the process of influence propagation into influence stage and reference stage,which correspond to the processes in which ordinary users become members and members become sticky members.Subsequently,this thesis also proposes Influence-Reference Rank(IRR)algorithm based on MBIC model.IRR quantifies the capabilities of nodes in different stages of MBIC model,and integrates these capabilities into one ranking index for selecting seed nodes by assigning appropriate weights.Experiments based on real-world datasets show that IRR is superior to classical algorithms and related basic algorithms in this field.(2)The problem of potential value based influence maximization with limited costFrom the perspective of merchants,this thesis explores how to maximize the potential value of customers using the tools of influence maximization.By analyzing the potential value of the user,this thesis decomposes the potential value modeling into two sub-problems of identifying potential customers and evaluating customer purchasing power,and proposes the problem of potential value based influence maximization with limited cost,namely the PVIMLC problem.In PVIMLC,nodes are given two attributes:cost and value.At the same time,the active nodes are further distinguished.Subsequently,three algorithms,Maximum Gain First(MGF),Maximum Gain-Cost Ratio First(MGCRF),and Maximum Revenue Expectation First(MREF)are proposed to solve PVIMLC problem.These algorithms evaluate the ability of nodes to obtain the potential value of surrounding nodes from different dimensions,and iteratively selects seed nodes based on the indicators obtained from the evaluation.Experiments show that compared with the classical algorithms in this field,all three algorithms proposed in this thesis have excellent performance.