Research On Message Propagation Prediction Model Based On Network Embeddin

The development of social network platforms such as Twitter and Sina Weibo has accelerated the generation and transmission of information,which may cause certain social impacts during the rapid spread of information.Therefore,accurately predicting the spread of news is a challenging problem,and it also has a wide range of applications in the fields of rumor suppression,viral marketing,recommendation systems,etc.Traditional feature-based methods rely heavily on manual feature extraction,which is cumbersome and difficult to be extended to other fields.Although the generation method considers the propagation mechanism,the prediction effect is poor.The method based on deep learning can realize end-to-end learning without the need of tedious feature engineering in traditional methods,and has good universality,so better information cascade prediction effect can be obtained.However,due to the many factors that affect accuracy of information propagation prediction,such as the structural characteristics of social networks,the dynamic changes of message propagation and so on,the prediction becomes more difficult.Based on these factors,this paper mainly proposes the following two information propagation prediction models based on network embedding.Firstly,this paper proposes an information propagation prediction model based on skip-gram called Skip Cas.First,the model uses the diffusion path and time effect at each diffusion time of the cascade graph to obtain the dynamic diffusion process of the cascade graph;Second,it uses the biased random walk to sample the cascade graph,and obtains the cascade structure features through the skip-gram model.Finally,the model combines the dynamic diffusion process and structure features to predict the propagation range of the information.Extensive experiments on Weibo and APS datasets show that the model significantly improves the prediction accuracy compared with the state-of-the-art methods.Secondly,this paper proposes a graph-level embedding based message propagation prediction model called HG-Cas.First,the model constructs a higher-order graph based on the similarity between different cascaded sub-graphs.Second,in order to make better use of higher-order neighborhood information of higher-order graphs,the multi-hop convolutional neural network is used to extract structural information so as to obtain the potential representation of nodes of different orders,and the attention mechanism is used to merge the potential representation of different orders to obtain the feature representation of nodes,and then the pooling mechanism is used to aggregate the node features and predict the propagation range of messages through MLP.Experiments on two public datasets,Weibo and Aminer,show that compared with the baseline methods,HG-Cas improves the accuracy of information cascade prediction significantly.Finally,based on the above two models,a prototype system of information propagation prediction based on Weibo platform is designed and implemented.The main functions of the system include: user information analysis,message forwarding analysis,message tracking prediction,and hot message prediction.