Research On Information Popularity Prediction Based On Social Network

With the increase of Internet penetration rate year by year and the maturity of social networks,social network has become an indispensable platform for information acquisition and dissemination.As one of the popular research directions of social network,popularity prediction is of great significance to enterprise marketing,public opinion research and judgment and other fields.The popularity prediction of social network information is analyzed and studied in this thesis,and the main work contents and innovations are as follows:(1)The methods of the information popularity prediction based on social networks are reviewed.Due to the lack of literature review related to the prediction of information popularity at present,the prediction methods of information popularity based on social network based on the full investigation of domestic and foreign literature is reviewed in this thesis.Firstly,the background and significance of information popularity prediction are summarized,the definition of information popularity is given,and the relevant theoretical mechanism is introduced.Secondly,the mainstream prediction methods are classified into three categories,and the basic principles of each method,the advanced research results at home and abroad and their adoption models are elaborated.Finally,The advantages and disadvantages of the three forecasting methods are summarized and compared.(2)A popularity prediction model based on feature engineering is proposed.In this thesis,a forecast model of information popularity based on feature engineering is proposed by analyzing the factors that affect the information popularity.The prediction problem is concretized into a binary classification task of whether the information popularity can exceed a specific threshold.An improved classification algorithm DCA is proposed by integrating logistic regression and random forest.Follow the basic process of feature engineering,analyze the factors affecting the popularity from three aspects of users,time and content,manually extract and construct the features related to the popularity,and further construct the machine learning classification model.Experimental results show that the proposed improved classification algorithm can effectively improve the overall classification accuracy,feature validity and importance analysis.(3)A popularity prediction model based on spatial temporal attention network is proposed.In order to capture the dependencies between the underlying structure and user interaction behaviors in the process of information transmission,the deep learning is used to learn the underlying semantics of cascading in an end-to-end manner,and a prediction model STA-Net based on spatial-temporal attention network is proposed in this thesis.The cascading sequence representation is obtained by using the sampling method based on the propagation path,the user representation in a specific scene is learned by embedding,the temporal dependence between nodes in the sequence is captured by the cyclic neural network,and the contribution of each node to the subsequent propagation is described by combining the attention mechanism.For each node,the feature representation of all its neighbors' fused timing information is gathered into itself to capture the structural correlation,and the multi-head attention mechanism is introduced to enhance the learning of structural information.Relevant experiments verify the necessity of each component of the model.Compared with the most advanced prediction methods,this method can significantly reduce the prediction error.The model is completely dependent on the timing and structure information in the cascade,so as to avoid large-scale and complex feature engineering,and it is universal for different popularity prediction scenarios.(4)A social network information popularity prediction system is designed and implemented.According to the above research,a social network information popularity prediction system is designed and implemented based on real social network platform data.The system can predict the prevalence of information and detect whether it is virus information.