Researches On Prediction Of Information Popularity For Social Networks

With the popularization of internet technology,online social network has become an important channel and carrier for people to exchange information in today's society.Information popularity prediction plays a vital role in online social networks,and there is extremely important research and application value.However,due to the diversity of information dissemination forms,the complexity of network structure and the multidimensional characteristics of user features,it is complicated and difficult to accurately control the information popularity.How to thoroughly analyze the changing regularities the popularity trend,how to perceive the trend of the topic popularity situation in advance,and how to establish an efficient public opinion control measures are the problems needed to be solved.The thesis mainly covers two aspects about information popularity prediction: macro information popularity prediction deeply analyzes the nonlinear dynamic mechanism for information popularity transmission and constructs a cross-platform information popularity fusion and prediction model.Micro information retweeting popularity prediction analyzes the micro factors affecting user retweeting and predicts the future retweeting information based on epidemic model.The detailed works of this thesis can be summed up as follows:1.At the macro level,the chaotic characteristics of social information dissemination are explored in depth,and the cross-platform information popularity fusion prediction model based on Bayesian estimation is proposed.Firstly,the information popularity time series from different social platforms are defined,and the principal components affecting popularity are quantified and obtained by Principal Component Analysis(PCA).Secondly,the chaotic characteristics of the popularity trend are explored,and the quantized sequences are reconstructed in phase space based on chaos theory.The changing regularities and properties of complex systems are restored in high-dimensional phase space.At the same time,the novel and fused phase space is obtained by using the Bayesian estimation to optimally fused multi-variable phase points in the same high-dimensional space.Finally,considering that the neural network has strong ability to approximate the nonlinear function,the neural network algorithm is applied to optimize and predict the fused information popularity.2.At the micro level,the multi-dimensional attributes affecting user retweeting are explored,and the infection rate in improved Susceptible-Infected-Recovered(SIR)model is quantized.A user behavior evolution strategy is proposed to perceive information popularity.Firstly,the retweeting driving force of user's personal and social dimensions is extracted.Further,the retweeting driving force is quantized by multiple linear regression.Secondly,in order to make the model closer to the real network communication architecture,the state S in traditional SIR is improved,and the propagation rules of SIR model are redefined.Finally,the state values of SIR model are extracted by time slicing technology,and Least Square(LS)method and quantified infection rate are used to fit the real model.A prediction method of information popularity based on user retweeting behavior and improved SIR model is obtained.In order to verify the validity and feasibility of the proposed method,the thesis validates the model based on real social network data set.Experiments show that the macro and micro information popularity prediction model proposed in the thesis can effectively perceive the information popularity transmission situation,and provide theoretical basis for macro situation analysis and micro user behavior analysis.