| ABSTRACT:With the continuous development of Internet and mobile communication technology,social network is expanding and the society has entered into the era of pan-socialization.Social network is not only an important platform for users to express their opinions and share information,but also a key carrier for media and enterprises to release news and push advertisements.The characteristics of openness,interactivity,personalization and intelligence of social network make the interaction mode and evolution process of users'opinions more complex and the factors that influence the formation of group opinions are more diverse.The development of the social network makes the way of information transmission between nodes transformed from one-way linear transmission to the interactive transmission,and also increase the user's participation.The reposting behavior,which prompt information to form the spread of two or more times,play a key role during the process of information transmission.The formation and development of network public opinion depend on the interactive behavior of users and the propagation of information.However,the traditional research of public opinion is not enough to find the key factors that influence the evolution of opinion and the propagation of information,and it is more difficult to predict and master the development of online public opinion.Consequently,we research on the problems of the interaction of individual opinion,the evolution process of group views,the mechanism of information propagation,reposting behavior and prediction by combining with the interdisciplinary research ideas and methods.The thesis mainly studies the influence of media heterogeneity on the interaction of individual opinion in social networks and explores the application of opinion dynamics model in practice.And the thesis makes an empirical analysis of users'reposting behavior,explores the reposting mechanism of information and the relationship between reposting data,and establishes a reposting prediction model.The research will be helpful to understand the process of formation and development of network public opinion,to enrich the content of the study of complex systems theory,and at the same time the research will have a certain guiding significance for the network public opinion control and Internet economy.The thesis's work has been supported by the National Natural Science Foundation of China(No.61271308,61401015),the Beijing Education Commission Application Research Projects,the Fundamental Research Funds for the Central Universities and the Beijing Key Laboratory of Communication and Information Systems.The main work and innovations of the thesis are as follows:1.Traditional model of opinion dynamics focuses on researching the heterogeneity of ordinary users,while ignoring the social network heterogeneity of different types of users.In the thesis,we study the characteristics of media users,and set up interactive evolution model based on heterogeneity of media.We use the modified RAS to model the process of the influence of media opinion to individual opinion,consider the influence of time factor on the efficiency of information transmission,time variable is adopted into the probability of opinion adoption to make opinion interaction model has timeliness.The results of simulation have shown that the characteristics with large average path length in rule network will lead to a phenomenon that the media opinion spreads slowly with a small scope,and the individual opinion interaction frequency is low,which leads to opinion evolution easy to fragmentate.On the other hand,the characteristics of strong heterogeneity and more center users in scale-free network lead to a phenomenon that the media opinion spreads fast with a big scope.And pinion evolution is easy to unified.Probability of opinion adoption has a huge impact for the final state average opinion value in the four kinds of network.Bounded confidence threshold has big influence on opinion evolution in the regular network and scale-free network.Media selection approach based on the node betweenness makes opinion influence transmission with high speed and large range compared with based on node degree and clustering coefficient.2.A lot of researches in the field of opinion dynamic focus on opinion update rules of model,the application research in practice is less.The thesis studies the application of opinion dynamics for advertising and marketing and sets up an opinion interaction model under the influence of the advertising.The model combines with the advantages of discrete opinion model and continuous opinion model.Meanwhile,it fully consideres the decreasing effect of the influence between friends and memory attenuation effect on individual opinion.The model defines the variables of advertising influence and advertising coverage,then takes research on the evolution process of group opinions and key factors under the influence of advertising.The results of numerical simulation show that advertising influence and advertising coverage have the similar effect for advertising.Only when the initial value is greater than a certain threshold,advertising can obtain significant effect.Advertising coverage plays the more important role than advertising influence in the same condition.Actually,the key factor for determining the success of advertising is the user activity,because user activity can directly reflect the degree of users' recognition for the product or service.This study will have certain guiding significance for the Internet economy.3.Reposting behavior is the key factor in the process of social network information diffusion.Traditional information transmission models lack of quantitative analysis of the reposting behavior,most of them are static description model.In the thesis,large amounts of microblogging reposting data has been analyzed and a novel dynamic information reposting model has been established according to the results of analyzing.The empirical analysis shows that:reposting is the main behavior when ordinary users use weibo;reposting depth obeys power-law distribution;more than 99%of the reposting behavior occurs within three times;reposting interval obeys exponential distribution.The thesis proposes a dynamic information reposting model based on markov process.The model uses the continuous time variable,which can make up the shortcomings of the discrete time information propagation model on prediction.Simulation results show that the proposed dynamic reposting model has better prediction ability compared with ARIMI model.In addition,reposting quantity is an important index for evaluating user influence,the model can also be used to solve the problem of maximum impact.4.Traditional repositng prediction model is not only not taking into account the relations between the reposting data but also lacking of the ability to handle huge amounts of information.In order to solve the problems,the thesis proposes a segmented BPNN model for prediction by using the advantage of artificial neural network in data processing.Firstly,analyzes the reposting data to find the characteristics of reposting behavior and establishes a segmented neural network model.By trial and error,the model determines the number of hidden layer neurons,the number of input layer,the learning efficiency and momentum factor and other parameters.The prediction results have shown that the segmented BPNN model has a more smooth prediction curve and the prediction accuracy has improved.Then,an improved segmented BPNN model has been proposed by using the efficiency of the genetic algorithm in global search optimal solution.The improved model not only gets rid of the inconvenience of trial and error parameters in BPNN but also avoids missing the best parameter selection.Simulation results prove that the proposed model enhances the accuracy for reposting prediction.This research topic provides a theoretical reference and practical guidance for future mass prediction. |
PDF Full Text Request