Relevance Modeling And Influence Discovery Of Multi Type Key Elements In Social Hotspots

At present,microblog,Twitter,Instagram and other major social platforms have become the main carriers of public access to information.With the rapid growth of information of users in social networks,the control of key elements(such as users and tweets)has become the core task of Internet public opinion control.Therefore,the influence discovery and influence prediction of key elements have important research significance.In this thesis,the ternary association graph model,knowledge representation,GCN and other methods are used to discover and predict the influence of key elements.The main work and contribution are as follows:1.In the aspect of key elements influence discovery,considering the relevance of messages,paths and users and in social network hot topics,and aiming at the complexity of feature space and relation space of multi types of key elements,this thesis proposes a method of multi types of key elements influence discovery based on ternary association graph and knowledge representation.Firstly,aiming at the relevance problem of multi types elements in social topics,this thesis constructs a ternary association graph model based on messages,path and users.Secondly,referring to the cross scoring strategy,this thesis proposes a key elements influence discovery algorithm based on ternary association graph and cross iteration scoring mechanism.Finally,in the process of cross iteration,knowledge representation is used to optimize the cross iteration scoring mechanism.2.In the aspect of key elements influence prediction,considering the diversity of relationships among multiple types of elements in social networks,and aiming at the complexity of social network with multiple types of relationships,this thesis proposes an influence prediction method of key elements in social topics based on knowledge representation and GCN.Firstly,according to the complexity of the relationship between elements,the user relationship network graph based on different paths is constructed.At the same time,according to the diversity of elements,the knowledge representation is used to learn the user embedding in complex space,and then the user relationship network graph based on multi types of elements and multi types of relationship spaces is constructed.Secondly,the user embedding in different networks is obtained by using GCN layer.Then,the final user embedding in different networks is realized by attention machine and is used to get user's influence prediction.In this thesis,real social network datasets are used to test the above model.Experimental results show that the proposed model can effectively mine and predict key elements influence.