Research On Heat Prediction Model Of Social Topics Based On Neural Network

With the continuous popularization and application of social software,the modern era has gradually become an era of the Internet,which has brought about tremendous changes in people's communication and lifestyle.By studying the heat trend of hot topics,it is easy to obtain the law of the generation and dissemination of hot topics.On the one hand,it can effectively understand the user forwarding behavior;on the other hand,it can analyze and supervise the network public opinion.It is an extremely valuable theoretical inquiry content.This thesis is based on the hot topic of social network,considering that the diversity of messages under the topic plays an important role in user participation behavior.The basic characteristics of users and the interaction factors of multiple messages are combined to study the user forwarding behavior from microscopic aspect and the heat of hot topics from macroscopic aspect.The following are the main research work and contributions of this thesis:1.Aiming at the influence of complex interactions between topic and multi-message on user forwarding behavior,a prediction model of user forwarding behavior under the influence of multi-message interaction is proposed and designed.Firstly,considering the influence of multi-message interaction on user forwarding behavior,based on the internal driving mechanism of user behavior,a multi-message interaction influence driving mechanism is proposed.Secondly,because the BP(Back Propagation)neural network is simple in structure and can respond well to the nonlinear relationship caused by the addition of multi-message interaction mechanism,it is used to construct the user forwarding behavior prediction model.Finally,because the multi-message interaction mechanism has an iterative guiding effect on user behavior,BP neural network is easy to fall into over-fitting situation.Therefore,the simulated annealing algorithm is introduced to optimize the traditional neural network to obtain optimal parameters,thus improving the accuracy of the algorithm.2.With regard to the heat trend of hot topics,the convolutional neural network and exponential smoothing method are used to predict the topic heat value of each time period,which can accurately and objectively describe the heat trend of microblog topics.Firstly,this thesis divides the topic into appropriate time periods.Secondly,in the microscopic aspect,the convolutional neural network is used to predict the user behavior,and the prediction results are added to approximate the heat value of the topic.In the macro aspect,the exponential smoothing method is used to predict the heat value of the topic in the next period.Finally,the two aspects are combined to perceive the overall heat of the topic.Finally,the experiment uses the actual data set of Sina Weibo to test.Experiments show that this thesis can not only predict the user forwarding behavior more realistically based on the actual situation of multi-message interaction,but also analyze the propagation law of the topic from a macroscopic perspective,so as to predict the heat trend of hot topics.It helps to analyze the scientific laws in the behavior change of netizens in hot topics,and provides theoretical basis for network public opinion guidance and complex marketing management decisions.