Events And Behaviors Prediction With Virtual Social Scenes

From the Internet to Mobile Internet,and now to the Internet of Things,we have been continuously digitizing and connecting the real world.We have also been exploring in this world,establishing rules of information,and creating a virtual world.It has also made the Internet itself gradually become the infrastructure of the virtual world and exist in the real world.The interweaving of virtual and reality has resulted in a new form of current society that is a fusion of the two.Faced with this new social pattern,it is urgent to study social events and behaviors prediction to solve social governance challenges and guide the sustainable development of society.Social events and behaviors prediction aims to use multi-source heterogeneous data to predict major events and specific behaviors that may occur in real society and virtual world,covering fields such as politics,economy and diplomacy.Breakthroughs in technologies such as artificial intelligence have made it possible to predict social events and behaviors.However,both the real world and the virtual world currently face challenges in data scarcity,dynamic evolution of social events and behaviors,and verification of prediction methods that may cause irreversible losses in real society if they are not accurate.Therefore,this dissertation proposes new approaches using game as a tool,combined with virtual social scenes,to solve the data scarcity,modeling and verification challenges in social events and behaviors prediction in the new social patterns.The main contributions of this dissertation are summarized as follows:(1)We propose a multi-level social event prediction method based on multi-source feature fusion to address the challenges of data scarcity and modeling in social events and behaviors prediction.Specifically,we conduct a hot topic attention research based on multidimensional features of social media content such as the average circulation and user attention,and propose a dataset of authoritative newspaper media in the target country.To address the problem of ignoring the multi-level characteristics of event evolution in existing social events prediction methods,we use large-scale pre-trained language model to classify the front page,define policy change index,and capture the policy change trend of the target country over a long period.We then propose a multi-dimensional siamese spatial and temporal dynamic graph network,which integrates event semantic features and spatiotemporal high-dimensional features to predict social events in a short period.Finally,we establish a multi-level conceptual model of “the long period leading short period,and the short period corroborating long period”.The effectiveness of our method is demonstrated through quantitative and qualitative experiments on self-constructed datasets and global event dataset.(2)We propose to use the sci-fi strategy mobile game Nova Empire,a typical virtual social scene in massively multiplayer online games,to solve the verification challenges in social events and behaviors prediction.However,the premise for using games to solve real social problems is to have a large number of online users as data support.Therefore,we reveal the virtual social engagement mechanism based on social patterns and behaviors.Using multilevel clustering to demonstrate that the distribution of players of different types follows a pyramid structure,similar to the distribution of social classes in the real society.Based on alliance and individual behaviors,we use correlation and regression analysis to reveal and verify the factors that affect game engagement in different stages.We focus on analyzing the relationship between different types of rewards and player game engagement in different stages,and demonstrate that the influence of random rewards on player online time is more significant than that of fixed rewards.By clarifying the mechanism of virtual social engagement,we provide theoretical support for using games to solve the verification challenges in social events and behaviors prediction.(3)There is currently no detailed and reasonable explanation for why massively multiplayer online games can be used to solve problems in fields such as sociology,economics,and artificial intelligence.Therefore,this dissertation proposes a bidirectional mapping relationship between virtual social scenes and real society based on multiplex networks.Specifically,using alliance behaviors to reveal the factors of social groups long-term survival based on the structure properties of virtual social networks.Social groups are the key to maintain the entire virtual social scene and are the basis of all researches.Based on multiplex network to model virtual social relationships,a multiplex network containing positive and negative virtual social behaviors is constructed.Furthermore,the structure properties of the multiplex network are analyzed,including reciprocity,node degree distribution,and smallworld characteristics.The social balance theory is verified in the virtual social scene by combining the principle of triadic closure.From the perspective of social patterns,we verify the structural consistency between the virtual social scene and real society,and propose a preliminary bidirectional mapping relationship between the two.Meanwhile,we provide a deeper theoretical support for the verification of social events and behaviors prediction methods based on games.(4)We propose a verification strategy based on virtual social scene to address the verification challenge of social events and behaviors prediction.The current mainstream verification methods based on test datasets ignore the complex evolution process and environmental information of events and behaviors.Therefore,this dissertation constructs a multiplex network that can represent complex virtual social behaviors,and proposes a node similarity metric method based on behavior polarity to predict events and behaviors in virtual social scenes.This demonstrates that events and behaviors in virtual scenes are also predictable,allowing games to be used for verification of prediction methods.From the perspectives of member distribution consistency,social structure consistency,and game mechanism consistency,we further analyze the principles and constraints of the bidirectional mapping.Finally,a verification scene and method are built based on Nova Empire,and quantitative experimental results are conducted to demonstrate the feasibility of the validation scene and the effectiveness of some classic prediction methods in real society.This dissertation not only addresses the challenges of data,modeling,and verification in predicting social events and behaviors in the real society,but also attempts to extend the conclusions obtained from games to real society based on the bidirectional mapping relationship between virtual world and real society.Specifically,this dissertation provides new solutions for precision marketing and sustainable society to game operators and policy makers respectively.The research of this dissertation not only has significant academic value,but also will provide technical support for China’s strategic planning,social governance,and sustainable development in the new social patterns.