Prediction And Analysis Of Collective Actions Based On Virtual Space

Real space witnesses the fact of social events,but it is difficult to provide relevant information to predict the state of social events.With the rapid development of the Internet,the virtual space based on the Internet has been basically formed.By participating in social events and sharing and disseminating information related to social events in the virtual space of the network,real individuals map social events into the virtual space of the network,thus gradually enhancing the correlation between the two.The network virtual space records the occurrence process of events.By using the relevant information in the network virtual space to quantify and analyze the correlation between the network virtual space and the real space,the development status of social events in the real space can be predicted,thus providing decision-making support for relevant departments.In order to solve this problem,this paper takes the development of social events represented by collective actions in virtual space and real space as the entry point and proposes a scalable integration information dissemination rule in time sequence model of the correlation analysis framework,the framework of people attention from social events,group emotional tendencies and events into time and probability to quantify,the analysis of the impact of network virtual space and real space,and is verified by events in the real space of development.This study is mainly carried out from the following aspects:(1)For the analysis of the influence of event attention,model-driven modeling is proposed from the perspective of information transmission law to build a differential equation model for the prediction of crowd attention.The model is reduced to a gray prediction model to solve the problem and realize the prediction of event crowd attention.(2)Based on the analysis of group emotion influence,a group emotion prediction model is constructed by integrating the law of population competition and the idea of infectious disease transmission,to realize the prediction of group emotion evolution,the explanation of formation process and simulation of event group emotion;(3)In view of the event into time and probability,based on the social media people attention,group emotions and events in the number of feature words,and the mainstream media reports,this paper builds training set with temporal characteristics for deep learning model from tag to feature that it can realize the occurrence time and probability prediction;(4)This paper proposes a semi-supervised method based on the generative adversarial network to realize the prediction of network events and solve the problems of insufficient data of some types of events and sample annotation.At the same time,the theoretical proof of the stable state of the enerative adversarial networks is improved.To verify the performance of the proposed models,the experiments verifies the four models based on Chinese microblog and Twitter data.The experimental results show that the performance of the four models in the prediction of collective actions is better than the representative researches.(1)The population attention prediction model proposed in this paper has a performance improvement of 11.17% compared with the existing researches;(2)The group sentiment prediction model has a performance improvement of 22.14% compared with the existing researches;(3)The optimal parameter configuration for the deep learning with the real data set is analyzed.The experimental results show that the proposed method is better than the existing representative methods from the view of precision,recall and F-measure;(4)For the semi-supervised method based on generative adversarial networks,the paper builds training set with a certain proportion of the generated data by the semi-supervised method and the real data.The experiment of predicting the occurrence probability of collective actions shows that the semi-supervised method based on generative adversarial networks is better than existing representative semi-supervised methods using different proportions of the generated data and real data.