The Research And Distributed Implementation Of User Influence Assessment In Social Network

Different from traditional networks, social networks such as microblog spread and diffuse information mainly depend on the multi-relationship between all users. How to make information received by as many users as possible, which is defined as the problem of influence maximization in social networks has become a research focus in social network platforms and its application. However, due to the huge number of users, traditional centralized data processing methods is restricted by their high cost and low efficiency.which makes the research of user influence not outstanding and leaves large quantities of data resource in the platform underutilized.PageRank algorithm is developed by Google to evaluate the page rank, which evaluate page rank by the situation of page chainout and chaininto. We abstract the relationship between users in social network into the relationship between the chainout and chaininto of pages and apply the algorithm to the analysis of user influence study, considering the influence of the followers at the same time. Based on which, we propose an improved algorithm URank to evaluate the user influence--URank algorithm.Particle swarm optimization algorithm is a kind of swarm intelligence algorithm, which attracts wide attention from researchers with the advantages of fast convergence and high precision and easy to realize. We propose a PSO-based user influence algorithm considering both user’s own and their folloewer’s influence. During the process, we evelautes user influence based on many elements such as the number and influence of fans, the frequency of information releasing, the forwarded or commented rate and whether whether the user is authenticated. Meanwhile, we determine whether the value of influence should be updated by fitness function which is the influence incremental factor. The improved algorithm ensures the reasonableness and fairness of the influence evelaution versatilely.In order to allowe the proposed algorithm to propose huge amounts of data, we adapt them into distributed parallel ones according to the MapReduce distributed parallel programming model. Our experiments are based on the real-world data crawled from SinaWeibo platform, which can reflect influence more fairly and reasonablely compared with other algorithms. Meanwhile, the analysis of clustering performance also shows that the parallel algorithm is of good speedup ratio and calculation efficiency ratio. In a word, the proposed distribution parallel algorithms are applicable for user influence evaluation in large-scale social netwok platforms.