Research On Prediction For Group Behavior Of Hotspots For Complex Feature Spaces

Due to the popularity of internet technology and advances in information technology,the principal method of information distribution has changed from offline to online social networks.Social networking sites have become the main channel for people to groups to exchange ideas and share information,for instance Sina Weibo and Twitter.Because of the independence,anonymity,and openness of social networks,once some news is published,it attracts numerous attentions and evolves into a hotspot that is commonly spread and discussed on the Internet.Therefore,the survey of topic dissemination and development trends in social networks is very important.The thesis researched topic dissemination on social networks from the micro and macro aspects.The retweeting and dissemination trends of the topic have been researched.The following essentials are the work of this thesis:1.Topic retweeting prediction based on single-message and user behavior.Firstly,in the hotspot,Generative Adversarial Networks(GAN)is introduced to generate homomorphic sequences.Secondly,with the complex feature space of topic propagation,this thesis proposes a representation learning method,called Topic2 vec,Topic2vec combines the influence of the node itself to set up a new walking strategy to mine hidden features and it uses an attention mechanism to improve the representation of features.Finally,considering the limitation of the active period of the topic,a topic retweeting prediction method based on Convolutional Neural Networks(CNN)is proposed by slicing the propagation period of the topic.2.Topic heat prediction based on multi-message and group behavior.Firstly,considering the role of different types of messages in the spread of rumor topics,this thesis draws on evolutionary game theory to quantify the interplay of different messages.Secondly,considering the heterogeneity and complexity of rumor topic propagation space,Knowledge Representation is used for mapping the knowledge triad of the topic dissemination network to the vector space.Finally,in view of the Non-station of the topic network,a group behavior prediction model based on the graph neural network,called KGraph SAGE,is proposed,which predicts the dissemination trends of the entire topic.Finally,the thesis chooses a real dataset of Sina Weibo to train and verify the model.Experimental results indicate that the model proposed can not only effectively predict user retweeting behavior on hotspot and group behavior in rumor topic,but also can predict the true spread of topics.