Research On Influence Propagation And Blocking Maximization In Competitive Networks

In recent years,with the development of era,the advancement of science and technology,the diversity of media,many large social networks such as Facebook,Twitter,Sina Weibo,WeChat,Tic Toc and other emerging media have been rapidly developed.The influence propagation in social networks has become the hotspot of research.In many social networks,individuals have positive and negative relationships,such as friends and enemies,cooperation and competition,agreement and opposition.This forms the different evaluations of the same item and the same viewpoint,resulting two opposite and competitive influences.In addition,the openness of social network platform leads to the spread of rumors,gossip and other types of misinformation,which will lead to a fast spread of negative information in the network.Negative and positive information,fake and true messages form a competitive relationship in the network.We call such network with two competitive influences as the competitive social network.In order to take advangtage of the social network,we must expand the propagation of positive information,and effectively block the spreading of the negative one.Therefore,this issue has an important application value for public opinion analysis,rumors control,epidemic tracking and control in the social network.However,current resaerch on the analysis of network influence propagation mainly focuses on the maximizing the propagation of positive influence.The study of blocking the negative influence is limited to the condition that the source of negative influence is known.Moreover,most of the existing methods use Monte Carlo simulation to estimate the influence propagation range,which can not be applied to large-scale networks due to its high computational cost.Therefore,based on the application background of network public opinion analysis,rumor control,epidemic tracking,etc,we study and analyze the current research of the influence propagation at home and abroad,and make an in-depth study on the maximization of positive influence propagation,the locating of negative influence propagation sources,and the maximization of negative influence blocking based on both deterministic and uncertain sources.The main innovations of our work in this paper are as follows:1.We studied the positive influence maximization method on signed networks based on the independent cascade model.We proposed a group of propagation rules of positive and negative influence on signed networks to establish a propagation model describing the propagation process of the two types of influences.Using the model,we can avoid the time-consuming Monte Carlo simulation.We proposed a propagation path-based algorithm to calculate the activation probability between the nodes.We defined an influence increment function,and proposed an algorithm to select the node with the largest influence increment as the optimal seed set.Experimental results on real social networks show that our algorithm can obtain a larger positive influence propagation range than other similar methods.2.We studied the influence sources locating algorithm based on representation learning on independent cascade model.We proposed an algorithm to calculate the activation probability between the nodes based on independent paths.We use the difference between the path length and the activation time of the observed node to measure the rationality of the nodeas the propagation source.We proposed the concepts of sending and receiving latentspaces,and map the network nodes to these two latentspaces respectively.We proposed an influence sources location algorithm based on representation learning.The algorithm computes the probability that the user will become a propagation source through his relative positions in the two spaces,and then determines the propagation source according to the probability computed.Experimental results show that the sources found by our algorithm can activate more observated nodes,and the activation time of observated nodes is more accurate than those by the other algorithms.3.We studied the issue of nagetive influence blocking maximization on the linear threshold propagation model based on the cohesive set.We proposed the concept of the maximum cohesive set under the linear threshold model,analyzed the basic properties of the cohesive set.We use the cohesive set to estimate the blocking effect of positive seed set on the negative influence.Using the cohesive set,we can avoid Monte Carlo simulation,decrease the search space for seed nodes,and reduce the time complexity.We proposed an algorithm to generate the cohesive set and a negative influence blocking maximization algorithm under the linear threshold model.Experimental results show that our algorithm can block more negative influence than other algorithms.4.We studied the issue of maximizing the blocking on negative influence from uncertain sources.We proposed a linear threshold propagation model of competition influence.Based on this model,we defined the concept of propagation tree on the live edge graph to estimate the influence propagation.Based on the propagation tree in the live edge graph,we proposed an algorithm to calculate the blocking increment of each positive seed.We proposed a node deletion-based algorithm to detect positive seeds to minimize the negative influence from uncertain sources.Our experimental results show that our algorithm can spread the positive influence to the maximum range and achieve more negative influence blocking than other methods.