Research On Influence Maximization Based On Moment Of Influence In Social Networks

The rapid rise of the Internet has pushed social networks such as Weibo and We Chat to gradually replace traditional media and become important media for interactive communication,information sharing and information transmission.As one of the hot researches on social network analysis,Influence Maximization(IM)Problem aims to select an influential set of seed nodes from social networks,which applies in many scenarios,such as marketing,rumor control,and epidemic monitoring.Therefore,with the rapid expansion of the social network and the difficulty in obtaining network structure,it has become a research hotspot to study these problems such as how to measure the influence of nodes or how to find an set of influential nodes to cause the maximum scale of influence propagation.And these problems are also the main focus of this paper.In recent years,researchers have proposed many influence diffusion models,node influence metrics,and influence maximization algorithms in response to the IM Problem.However,there are some limitations.On the one hand,the performance of existing influence metrics based on structural characteristics of node varies greatly in different networks.On the other hand,the existing influence maximization algorithms perform poorly in considering efficiency,accuracy and scalability overall.In addition,most studies assume that the topology of the network is completely known,while the research on the IM Problem under partially observable networks is still immature.To overcome the above shortcomings,this paper is dedicated to the research of the IM Problem in social networks from two critical perspectives: node influencemetrics and influence maximization algorithms.It is embodied in the following three aspects:(1)In order to quantify the influence of nodes effectively and explicitly,this paper is inspired by the torque in automobile dynamics to define Moment of influence(Mo I),an inclusive and flexible influence metric.It is a power form of local diffusion of node influence,which could reflect better the essential characteristics of node influence.The effectiveness and robustness of Mo I are verified by experiments.(2)Random Nearest Neighbor Recommendation(RNNR)algorithm based on Mo I is proposed.The RNNR algorithm is divided into two stages,(i)initial nodes selection by "Random Selection";(ii)determining the seed set by "Nearest Neighbor Recommendation".Aim to avoid the influence overlap effectively,we preprocess the network with the community structure,so as to optimize the initial node layout.The algorithm complexity isO(?),which is almost independent of the number of nodes or edges in the network,so the algorithm is suitable for large-scale network and has good scalability.Finally,the effectiveness and scalability of RNNR are proved by both theory and experiment.(3)A solution to solve the IM problem under partially observable network is proposed.Firstly,the definition of partially observable network is given.Considering the limitation of unknown network structure,DPSO is introduced into RNNR algorithm to optimize the initial node layout,and Nearest Neighbor Recommendation Based on DPSO(DRNNR)is proposed.Experimental simulation shows the effectiveness of the algorithm.