Study On Influence Maximization And Diffusion Probability Representation And Prediction Based On Social Networks

With the development of information technology and the demand of communication,many well-known large-scale websites have emerged,such as Weibo,We Chat,Facebook and Twitter.These websites have been widely used in many aspects,such as social network s,information dissemination and influence diffusion.Research in recent years reflect that people are more willing to accept product recommendations from family members or close friends than from public channels such as leaflets and newspapers.This so-called “word of mouth” social phenomenon shows that we can make full use of this phenomenon to carry out targeted information delivery,in order to obtain a broader dissemination effect.As an important research content of viral marketing,this problem is summed up as the problem of influence maximization.Based on the problem of influence maximization,many branches have been formed during years of research.This paper focuses on two aspects: firstly,the problem of budgeted influence maximization with each node in networks has different social influence and different activation cost.Furthermore,the problem of reasonable representation and prediction of the diffusion probability between nodes is discussed.Around the above two research directions,this paper has carried out in-depth analysis and research.The main work contents are as follows: 1.As the existing research methods focus on the problem of influence maximization itself,the solutions to the problem of budgeted influence maximization with different node activation cost mainly focus on the improvement of greedy method.Therefore,based on the analysis of the relationship between nodes in the network,this paper proposes a new method of maximizing the influence of the budget based on the simulated annealing algorithm.The algorithm Boost SA can effectively solve the problem of high computational complexity caused by the dependence of greedy method on Monte Carlo simulation,and can achieve better performance improvement in almost the same computing time compared with the same type of algorithm Combination SA.2.Due to the lack of effective feature representation and prediction for the propagation probability between nodes,most of the existing research methods only assume that it is a fixed value or a simple distribution,which deviates from the actual scene.Although the learning method of node embedding representation can indirectly realize the edge vectorization representation through the node vectorization representation,there is inevitably a certain degree of information loss.Therefore,we propose the Combination algorithm,which improves the existing edge embedding representation method line2 vec and integrates some known propagation information to achieve a reasonable representation of the existing network edge.Through the combination of the two representations,it is applied to the effective prediction of the propagation probability between nodes.In the actual networks,experimental results show that the Combination method has a better improvement than the original method.3.In the process of using representation learning method to predict propagation probability,the existing method Deep Walk thinks that the known propagation sequence can be directly used to replace the node sequence generated by random walk step to realize the vector representation of nodes.In view of the shortcomings of this method,we pro pose an intuitive RBC algorithm based on the reconstructed subgraph and an improved BRBC algorithm based on the reconstructed complete network sketch to solve the problem of running time bottleneck of RBC algorithm,so as to realize the reasonable generation of node sequence.The experimental results show that the BRBC algorithm has a strong advantage in computing time compared with the RBC algorithm,and reasonably solves the shortcomings of Deep Walk.