Research On Information Propagation Prediction In Social Network

With the development of technology,online social networks have risen rapidly.Everyone in the social network can participate in the process of information propagation,form a huge information transmission network,which enables highly integration of user relationships and information interaction.However,information dissemination becomes intricate and difficult to control because of the complexity of online social network topology structure and the multidimensionality of user behavior characteristics.It can not wait to do some research about ascertaining the information dissemination rule of online social network,predicting propagation behavior and trend,excavating the key transmission users,and setting up effective preventive measures.This thesis of information propagation prediction mainly focuses on two aspects.For the prediction on nodes,the user is center,the influence factors are studied to predict user forwarding behavior.For the prediction on structure,the information is center,the trend and scale of information dissemination are studied to predict information popularity degree.The main contents of this thesis are listed as follows:1.In the aspect of node prediction,a forwarding prediction model is constructed based on user behavior and relationship data.First,aiming at the complex causes of forwarding behavior,this thesis proposes three driving mechanisms,including interest-driven,habit-driven and structure-driven.Second,due to the multidimensionality of user features,by taking advantage of the Latent Dirichlet Allocation(LDA)model in dealing with problems of polysemy and synonymy,the traditional text modeling method is migrated to user feature modeling to accurately mine the multidimensional user features.Meanwhile,in regard to the modeling of continuous attribute values,the thesis used a Gauss distribution to improve LDA and proposed a novel forwarding prediction model based on improved LDA.Finally,considering the impact of timeliness,time discretization method is used to dynamically monitor user feature,thereby the prediction precision can be improved.Experimental results show that the proposed model not only can predict forwarding behavior effectively,but also mine the key factor that affects user forwarding behavior.2.In the aspect of structure prediction,a popularity prediction model is proposed leveraging node behavior data and information propagation networks in social network.On the one hand,taking into account the difference of users and the difference of relationship strength,the influence node's own and the influence of node pairs are introduced from the individual behavior dimension and the node interaction dimension.Then,an improved PageRank algorithm is designed to measure the node transmission ability.On the other hand,considering information dissemination is mainly affected by original publisher and earlier forwarding group,a popularity prediction model is proposed based on node transmission ability by Logistic Regression(LR)classifier,combining the individual features of publisher and early forwarding features as model input.Experiments show that our model has better ability to predict the popularity degree of information,and it also can discover network group events as well as identify important nodes in propagation networks.