Microblog Forwarding Prediction System Based On Deep Learning

Social media is one of the rapidly growing internet sectors in recent years,with a large number of users sharing information and distributing status on social media networks every day.As the earliest developing social media network platform,microblog has attracted a large number of users to register and use.Forwarding is an important mechanism for microblog to realize information dissemination.After receiving information shared by other users,users can forward the information to their own social media platform to share information to more users.The rise of the mobile Internet,such as 4G networks,large-scale coverage of Wi-Fi networks,and the rapid spread of smart phones,further attract more users to use microblog.Based on such an application scenario,it is of extraordinary significance to analyze the microblog forwarding process.For companies,predicting the number of microblog forwarding can help companies to monitor traffic and adjust devices to improve the user experience.For the government,an important part of the public opinion analysis is to try to get the focus of current social media users.To a certain extent,microblog forwarding prediction can help the government cope with group events caused by the explosive spread of information.This paper mainly describes the characteristics of the microblog forwarding process from three perspectives,namely user social network structure,user forwarding microblog time interval and natural language features of the microblog text,and uses deep learning framework to predict the forwarding amount of a given microblog.The research content of this thesis is mainly composed of two parts: one is to extract the network structure features of each user by using the graph embedding method in the social media network structure of a given user.The basic idea is to use the graph vector composed of nodes and connected edges,to achieve the purpose of dimensionality reduction and information extraction.The characteristics of the network structure can reflect the characteristics of the online social community where the user is located(such as whether the connection between community users is close,the similarity among the users),thus providing information for microblog forwarding prediction;the second is using recurrent neural network.The network integrates the user's network characteristics,the time interval and regularity of the user's forwarding microblog,and the natural language features of the microblog text,and obtains the forwarding prediction feature of a deep learning model for a given microblog.The obtained forwarding prediction feature is used to predict the number of microblog forwarding.This paper proposes a Fast Line graph embedding algorithm based on first-order similarity and second-order similarity of social networks and a microblog forward prediction system based on recurrent neural network.Both Fast Line and microblog prediction systems are the most advanced in recent years.The advantages of the algorithm is the improvement of efficiency.The Fast Line graph embedding algorithm of this thesis uses the method of approximating the objective equation to make the parameter training speed faster than the baseline method,and the accuracy of the graph embedding result is the same as the baseline method.The microblog prediction system of this paper can improve the prediction accuracy by 30% compared with the prediction results of the DeepCas method.