Research On Social Network Information Cascade Prediction Based On Graph Convolution

With the rise of social networks,social sensations caused by the explosive reposting and sharing of tweets have become commonplace.The phenomenon caused by forwarding and sharing is also called the information cascade phenomenon.The cascading dissemination of information forms an immeasurable social force and has a profound impact on society.In this context,it has important theoretical significance and practical value to predict the cascade of information by studying the process of information dissemination.On the basis of consulting a large number of domestic and foreign related documents,this paper comprehensively uses generative modeling and deep learning methods,and builds an information cascade prediction model in social networks through reasonable coding of the information dissemination process,and conducts an empirical analysis of the model.The main work content and innovations of this paper are as follows:1)A cascade prediction model of social network information based on spatiotemporal attention is proposed.In order to capture the dependence between user interaction behaviors in the process of information dissemination,this paper proposes a social network information cascade prediction based on spatio-temporal attention by explicitly learning the representation characteristics of temporal information and spatial structure information in cascaded information(ICP)model.In order to obtain the spatial structure information of the cascading information,the graph convolutional network is used to learn and propagate the representation characteristics of the cascaded graph,and the characteristics of its neighbor nodes are aggregated to its own node,and the two-way cyclic neural network is used to learn the timing between nodes in the cascaded subgraph.Information and interaction,combined with the attention mechanism to couple the spatial structure information and time sequence information in the calculation process.Finally,the verification on two real cascaded data sets,the experimental results prove the necessity of each part of the ICP model and the high efficiency of each variant.Compared with the previous cascade prediction model,the prediction error of this model is significantly reduced,and it has good versatility in different prediction scenarios.2)A self-excited point process graph convolution cascade prediction model is proposed.This paper considers the point process of information dissemination,uses graph convolutional network,random walk,and self-excitation mechanism to propose a selfexcited point process graph convolution cascade prediction(GHawkes)model.First,encode the user's influence according to the feature information of neighbor nodes,use random walks for sampling,and then use the graph convolutional network to perform graph representation learning on the user's influence,learn the relationship between users,and then go through the self-stimulation process The dissemination structure of cascaded information is learned,and then the attention mechanism is used to couple user influence and selfmotivation process,and output prediction through content attenuation.Finally,the verification on two real cascaded data sets shows that the experimental results show the necessity of each part of GHawkes and the efficiency of each variant;compared with the previous prediction model,the prediction error of this model is significantly reduced,and the difference is different.All prediction scenarios have good versatility.