Research On Community-based Competitive Influence Maximization Algorithm In Social Networks

The emergence of online social network has subverted people’s traditional social mode and is gradually becoming an important platform for obtaining and sharing information.Hot issues such as intelligent recommendation and online marketing are almost everywhere in people’s production and life.By defining such problems as influence maximization problems,researchers are interested in exploring how to solve these problems efficiently.However,with the rapid increase of data scale,the traditional influence maximization algorithm has shown its limitations in dealing with large data sets,and has been unable to meet the new computing standards and processing requirements.Therefore,how to obtain the maximum range of active users efficiently and accurately has become the focus of research at home and abroad.After studying the influence maximization algorithms at present,this thesis finds that the influence maximization methods are limited to the transmission of messages in a single channel,but ignore the diffusion of messages in multiple channels with competing relationships.Therefore,this thesis introduces local similarity,formulates a new competition and propagation mechanism,and simulates the real network environment to increase the accuracy of calculation.The main research contents of this thesis include:(1)In the stage of community discovery,local similarity is introduced in this thesis and a label propagation algorithm based on local similarity in non-overlapping communities is proposed to solve the problem that there are a lot of low-quality and redundant data in social networks and the recognition accuracy of existing label propagation algorithms is low and unstable.The algorithm screened high-quality posts and high-quality users based on HITS algorithm,then used the complete subgraph algorithm based on local similarity to optimize the initial propagation,and defined the influence of nodes in the propagation process to identify the target community more quickly and accurately.Finally,through the comparative experiment,it is verified that compared with the traditional label propagation algorithm,the accuracy stability of community identification is greatly improved.(2)In the stage of influence maximization,this thesis introduces the concept of competition in view of the lack of simulation of the real environment of social networks in existing studies,and puts forward the algorithm of influence maximization in the competitive environment according to its characteristics.By comparing with other efficient algorithms,this thesis firstly puts forward the influence propagation model based on competition,and then puts forward the node model based on user interest,including the solution of user interest distribution,the calculation method of influence probability between nodes and the comprehensive influence of nodes.Finally,we prove through experiments that the influence maximization algorithm has a certain improvement in the scope of influence,relatively small time complexity,and a significant improvement compared with other algorithms.