Reseaerch Of Information Heat Prediction Based On Social Networks

Social networks information heat prediction has been one of the most important research content in information content security administration field as the internet is rapidly developing.In before information heat prediction research,the research of the information representation and quantization calculation is not sufficient enough,so it is difficult to effectively support the modeling and prediction.Meanwhile,because most of the current social networks dissemination model doesn't consider memory effect in the process of the information dissemination,there is great deviation between the predicted results and the actual results.Firstly,using na?ve Bayes text classification algorithm to divide the collected blog text into fifteen fields.Then,we establish a scientific reasonable index system tree and use expert scoring method and Analytic Hierarchy Process(AHP)method to assign weights to all levels of indicators in order to analysis user influence in various fields and comprehensive influence.Next,in the process of the social networks information heat prediction,we consider both interest accumulation and aging attenuation about memory effect.Then,using genetic algorithm to obtain memory curve in order to analysis the influence of the users' memory effect on social network information dissemination.Meanwhile,we obtain strong and weak tie users about the home page or users in the social network and analysis the relation between weak tie users and information transmission,and then we get the impact of the weak tie user on the information propagation scope.Finally,we analysis the scope,trends and speed of the information dissemination quantitatively for the sake of revealing the influence of user characteristics and content relevance on the user behavior in the social network.We consider specific topic and post two levels,and then adopt machine learning algorithm GBDT(Gradient Boosting Decision Tree)to establish the prediction model for getting the scope and trends of the information propagation.In our experiment,according to single variable principle,it is compared that the impact of the different features on information propagation heat prediction.Experiment results show that higher accuracy can be obtained by means of considering user influence,weak tie,memory effect,repost number,comment number and like number synthetically in addition to the posts that the heat values are less than 10,and the average MAPE value is well below 30.