Multi-topic Competition Propagation Modeling And Forecasting

Social networks play a very important role in people’s lives.Based on the social network information dissemination law to predict how hot of a message,it will help users to grasp the macro situation of network public opinion,and guide the publication and evaluation of network information.At present,many researchers have devoted themselves to discovering the law of information diffusion,and many important achievements have also been made in the field of information diffusion.However,many studies focused on how single message is spread in the network,for the multivariate analysis of diffusion of information about the mutual interference competition diffusion model is still in the primary stage,the related research work is little.This paper focuses on the propagation of multiple topics,by introducing the topic interaction matrix and user interaction matrix in SH model,proposes a multi-information oriented competition complex opinion environment hot topic prediction model can objectively reflect the regularity of information dissemination in actual network.At present,partial diffusion model takes into account that many messages are transmitted in the network at the same time,but they ignore the relationship between users and information.At the same time,most of the previous studies use implicit feature modeling,which makes it difficult to combine reality with the interpretation of the law of information dissemination.In this paper,the method of probe users is added to the interaction matrix between users and information,so that the accuracy and recall rate of single user prediction tasks has improved.The traditional models are generally divided into two types;early hotness using a single information,release time characteristics of single information modeling;each time the hotness,predictive modeling time series model.The common problem of these two methods is that they do not take into account the interaction of many news information,so the effect is not very good.The main work of this paper includes the following aspects: 1.The existing algorithms do not consider the degree of closeness between the forwarding and the source of social interaction,which is an important factor affecting the scope of communication.On the basis of the strong correlation feature of the traditional prediction model SH,we add information early release time and strong connection user proportion characteristics.2.Most of the existing algorithms use a single model such as linear regression.The model is too simple or prone to overfitting.This paper uses multi stack model(Stacking)and fusion(Bagging)to improve the accuracy and avoid over fitting phenomenon occurred.3.Based on the prediction model with interaction relationship between categories,while also adding users and categories of interaction,finally get the response categories,the interaction between relation matrix,not only a part of the explanation for the dissemination of information on the law,and effectively improve the prediction results.The experiment was carried out on the Tencent news data set.The experimental results show that compared with the traditional SH model,the root mean squared error is reduced by 35%,the mean absolute percentage error decreased by 30%.