Research On Propagation Prediction Of Rumor And Anti-rumor Game Based On Sparse Representation

With the advent of the era of big data,the Internet has gradually become a ubiquitous way to obtain,share and spread information.The online social networking platform based on the Internet has become the main way for people to communicate.However,social network not only brings convenience to people's production and life,but also brings many negative harms,such as online rumors.Compared with traditional mainstream media,the freedom and rapidity of online social media make the generation of online rumors easier,spread faster and do greater harm to the society.Therefore,it is of great significance to effectively predict the spread trend of online rumors for public opinion monitoring.This thesis focuses on two levels of single-message individual behavior and multi-message group behavior under rumor topics to carry on the prediction to the rumor topic spreading.The main contributions of this thesis include the following two points:1.At the level of single-message individual behavior of rumor topic.Firstly,considering the advantage that sparse representation can represent all original samples with as few atoms as possible in sample representation,the sparse representation algorithm is used to vectorize the rumor propagation space with low rank.Then,the tensor completion algorithm is used to compensate the spatial data of the rumor topic.At the same time,the time decay function is introduced to quantify the dynamic behavior of users in the process of rumor propagation more truly.Finally,considering that rumor and anti-rumor present a cooperative and competitive relationship when rumor topic is propagated,a user participation behavior prediction model based on competitive cooperative graph convolutional neural network(CC-GCN)is designed.2.At the level of multi-message group behavior of rumor topic Firstly,the user attribute characteristics and multi-message characteristics are extracted from the perspectives of individual user factors and multi-message factors that affect user behavior.Secondly,aiming at the game relationship of multiple messages in the rumor topic space,this thesis quantifies the driving forces of multiple messages with the help of evolutionary game theory,reconstructs the rumor dynamic transmission network,and uses the representation learning method to extract the user structure characteristics under the dynamic transmission network.Finally,considering the advantages of multi-model fusion in improving the generalization ability of the model,the individual attribute driven submodel and the multi-message driven sub-model are fused to construct the group behavior prediction model of multi-message interaction affecting.Finally,experiments and analysis are carried out on the data set of Sina weibo platform.Experiments show that the model proposed in this thesis can truly reflect the game relationship between multiple types of messages in the process of dissemination,effectively predict the user's participation behavior,accurately fit the rumor propagation trend,and accurately explore the rumor propagation rules.