Propagation Prediction Of Text Rumors In Social Network Based On Representation Learning

With the advent of the information age,social networks have become a platform for people to quickly acquire and share information,but also provide a hotbed for the rumors propagation.The rumors have fast propagation and great lethality.It is crucial for controlling the rumors to predict the rumors propagation effectively.In recent years,many researchers have put forward various methods for the prediction of rumors propagation.These methods have good results,but at the same time they also have various deficiencies.This paper summarizes the advantages and disadvantages of the existing methods,and proposes a prediction method based on the representation learning and the deep neural networks.The main work of this paper is as follows:1)A representation learning method of user's social content features is improved.The social content posted by the user is treated as an 'article' and keyword sequences in the article are extracted to remove the noise of the social content and only others showing the user's main social interest are retained.Then the doc2vec algorithm is used to learn the user's social content feature representation through the keyword sequences.Finally,a user's social content features can be represented by a low-dimensional dense real-valued vector.2)A representation learning method of user's social structure features is improved to simultaneously learn the explicit attention relationship structure features and the implicit semantic relationship structure feature.Firstly,the social network and the knowledge graph are connected by the user-entity association relationship.Then,the degree of association between the user nodes is calculated using the KI(Katz Index)coefficient,and a new social network adjacency matrix is obtained.Finally,the node2vec algorithm is used to learn the user's social structure features.In the end,a user's social structure features can be represented by a low-dense real-valued vector.3)A variety of artificial features are designed to represent the user's social behavior,including the social behavior frequency,time distribution and interaction features with neighbor users.Afterwards,the Stacked Auto Encoder is used to learn the user's artificial feature representation.Finally,the encoded low-dimensional vector is learned to represent the user's social behavior.4)A rumor propagation path prediction model based on neural network is proposed.The standard convolutional neural network((CNN)cannot effectively express the structural features of topological graphs such as social network and knowledge graph.This paper uses a CNN variant,graph convolutional network(GCN),to utilize the propagation path prediction of network rumors.The user's content embedding and behavior embedding are concatenated as the user feature vector,which is as the input of GCN.The GCN-based model can predict the unknown users' propagation behaviors with a part of users that have propagated the rumors.The Experiments show that the prediction propagation model based on representation learning and deep neural network proposed in this paper can predict the propagation path of network rumors effectively,and has a significant improvement in performance compared with the traditional methods.It has a good application value for effective prevention of rumors in the current social environment.