Content-Aware Social Influence Modeling And Predicting

Social influence is almost everywhere around us.It not only appears in our real life,but also exists in virtual cyberspace.It is a very important research topic in social network analysis.Generally speaking,social influence refers to the phenomenon that a person's emotions,opinions,or behaviors are influenced by others to produce certain behaviors.Traditional social influence prediction methods usually design some manual rules to extract the characteristics of users and networks about a particular function,so the effectiveness of these methods depends to a large extent on the domain knowledge of experts,and these methods are difficult to correctly estimate the influence parameters of the edge in the absence of observed data.Benefiting from the emergence of the network representation learning method,it is a very convenient,fast and effective method to automatically extract the characteristics of nodes and networks using the network representation learning method.These methods convert nodes in the network into a low-dimensional continuous vector representation by learning a feature mapping function.Currently,related algorithms based on network representation learning methods for social influence prediction have been proposed.They take advantage of the user's structural information,attribute information,and neighbor's historical behavior information when modeling,but they cannot effectively process content information,that is,these model cannot explain what content the user will behave.Social media content information contains a wealth of knowledge,which can reflect the user's interests,research directions and other information.Modeling with social media content information can not only enhance the effectiveness of the results to a certain extent,but also make the prediction results interpretable.In order to solve the above problems,this paper proposes a content-aware social influence modeling and predicting algorithm.The main work is as follows:1.In view of the problem that the existing models cannot effectively use content information,so that the prediction results cannot be explained,this paper proposes a content-aware social influence modeling and predicting algorithm.This algorithm can not only predict whether the user is influenced by the neighbors,but also know which socialcontent the user will behave and enhance the interpretability of the prediction result.2.This paper implements content-aware social influence predicting algorithm.Based on the current popular ‘pytorch' deep learning framework,we use graph convolutional networks to represent nodes,and long short-term memory represent social media content.The two representations are connected through a relational network to predict whether the user will behave on specific social media content.Experiments and analysis are carried out with the latest social influence prediction method,the results show the proposed algorithm outperforms the comparison algorithm in precision,recall and F1.