Research On The Positive Influence Maximization In Signed Social Networks

With the constant innovation and development of Internet technology,diversified social platform have become one of the important ways of information dissemination,such as Facebook,Twitter,Weibo,etc.An increasing number of users interact by sharing their information on social media.How to rapidly spread the information shared by users and make influence wider,which has become a research hotspot in the field of social network analysis.Social media is a tool and platform for social interaction,providing technical support for users to communicate,discuss,exchange and share information.Users in social media have dual roles:they are both producers of information release and receivers of information processing.At present,influence maximization has tremendous application value in marketing,product recommendation,rumor control and so on.Signed network is a social network that can reflect both positive and negative relationships between users,which can accurately simulate the real social process.However,the current research on influence maximization is based on unsigned network,which leads to inaccurate influence evaluation.This paper studied positive influence from the perspectives of propagation model and seed set selection strategy,propose novel models and improved greedy algorithm to solve influence maximization.Finally,extensive experiments prove the rationality and effectiveness of the proposed models and algorithm.The main research content and innovation of this paper is as follows:1.The concept of net positive influence is introduced and the problem of net positive influence maximization(NPIM)in symbolic networks is proposed to select a seed set with as much positive influence as possible and as little negative influence as possible.An improved R-Greedy algorithm is proposed to solve the NPIM problem.Experiments on Epinions and Slashdot data sets show that positive influence and net positive influence are different in terms of concept and solution method,which also indicates that the solution proposed in this paper can achieve better diffusion effect in a shorter running time.2.To select the seed node set with the largest spread of positive influence within the specified Time,Time-sensitive Positive Influence Maximization(TP-IM)problem is proposed.In addition,a polar influence diffusion(HDPID)model based on thermal diffusion and an improved K-step greedy algorithm are constructed to select seed node sets to solve the TP-IM problem.Experimental results on three symbolic network datasets,Epinions,Slashdot,and Wikipedia,show that the proposed method in this paper shoes better performance in positive influence propagation with time limit.3.RandomWalk in DeepWalk model is studied and improved.The polarity factor is integrated in the process of generating sequence,and the topological structure of network and local neighborhood structure of nodes are extracted at the same time,so that the vector representation can more reflect the characteristic information of nodes in symbolic network.Experimental results on Slashdot and Epinions data sets show that better diffusion results can be obtained by using the proposed method.