A Research Of Quantifcation Of Pairwise Infuence In Online Social Network

Recently, due to the explosive growth of the large online social network, it resultsin a large number of applications, the most signifcant application of which is viralmarketing. The core issue is how to quantify users’ infuence on online social network.Currently, the main method is to break down the process of infuence quantifcationinto two stages, in the frst stage, we quantify the pairwise infuence, then in the secondstage, we quantify users’ infuence based on their pairwise infuences. This paper ismainly focus on the model of pairwise infuence quantifcation.Existing pairwise infuence quantifcation models are focus on traditional socialnetworks, and therefore does not apply to online social networks. Although the re-searchers also proposed some pairwise infuence quantifcation models specifc to on-line social networks, but these models exist some shortcomings, which may lead toinaccurate quantitative results. Since the quantifcation of pairwise infuence is thefoundation of the entire application, which will directly afect the accuracy of fnalresults, we need a model which can quantify the pairwise infuence on online socialnetwork accurately.In this paper, we choose the micro-blogging platform and defne the pairwise in-fuenceasretweetprobability. Firstly, thispaperdeducesuniformmodelandmaximumlikelihood model theoretically and analyze their shortcomings. Subsequently, we ver-ify the results of the theoretical analysis through real dataset. Then, this paper revisethe existing models by introducing the Bayesian probability theory and the concept ofdummy retweet into the existing model and propose the Bayesian model of pairwiseinfuence quantifcation. Ultimately, we prove that the uniform model as well as max-imum likelihood model are the special cases of the Bayesian model. In the experimental section, this paper verifes the theory of the model directlyand indirectly. We prove that Bayesian model solves the shortcoming of maximummodel and provide a more accurate pairwise infuence quantifcation through compar-ing the pairwise infuence quantifed by the models directly. In the meanwhile, we alsoprovethatBayesianisageneralizedmodelandprovideamorecomprehensivepairwiseinfuence quantifcation through both qualitatively and quantitatively analyzing the in-fuencers list found by diferent models.In the end, we analyze the core role played bydummy retweet in the Bayesian model.