Analysis And Mining Research On Influence Propagation In Social Network

Social networks have become an important vehicle for the public to share and exchange information. Influence propagation is one of the important characteristics of social networks. To analyze and mine the influence propagation benefits for information diffusion, product marketing, advertising, and public opinion control and so on. However, with the development of network and demand of users, the different importance of benefits of different information embracers to the promulgator, existing studies are still difficult to meet users’need. As a result, this paper focused on the research of influence propagation problem in social networks by using users’behavior information and social relationships, and etc. The specific researches and the main results as follows:First, for the influence quantization problem between users in social networks, according to the accumulation characteristic in the course of influence propagation, this paper proposed an influence weights learning method between users in social networks based on linear threshold model. We first estimated the probability density function of user’s activated threshold value by using the maximum entropy principle, and calculated the probability that the user is activated based on the probability density function. Then from the thought of maximum likelihood estimation, the influence quantization problem between users on the basis of user’s history behavior in social networks can be modeled as an optimization problemunder certain constraints and according to the characteristics of the target function and constraints, the corresponding algorithm was proposed.The algorithm can effectively learn the influence between users by optimization strategies on the basis of particle swarm method, including problem mapping, fitness function establishment, cross-border block, dynamic parameter setup, optimal particle variation and etc. Finally, the experiments verified the effectiveness of the proposed method with the real social network datasets and relevant user behavior log.Second, for the influence maximization problem, according to the social relationship and user’s historybehavior information, this paper proposed an effectiveinfluence maximization method based on the interest community division. We first took advantage of both the user behavior and social relations to measure the user similarity, and based on which divided the social network into several interest communities. Then with the dynamic filter boundary values, the influential nodes with maximal influence marginal benefits were chosen out from these communities by greedy strategy. Finally, the experimental results showed that this method can improve the working efficiency with approximately the sameinfluence effect.Third, for the target-oriented influence maximization problem, according to the influence propagation model, this paper first modeled this problem with objective function and designed random function to simulate the objective function. This random function calculated the influence from other to the target with the characteristics of propagation of influencein two steps, which was proved to have lower variance. Then as the sub-modular of objective function, an approximate algorithm was proposed with greedy strategy. Furthermore, an efficient heuristic algorithm for large-scale networks was also proposed to solve this problem. Finally, the experimental results based on real social network datasets verified the effectiveness of the proposed methods.