Topic Trend Prediction Method Based On Deep Learning

In recent years,with the development of the Internet,more and more netizens have participated in the discussion of social hot issues through the internet,especially for social sudden sensitive events,and more and more people have participated in them,by expressing their opinions through various channels,netizens' discussion of hot social topics may become an online public opinion event if it is not controlled.How to correctly predict the trend of topics on the Internet,conduct in-depth research on the law of information dissemination and find effective methods for predicting the trend of topics can help to control the occurrence of public opinion events,and became a pressing issue.The main work of this paper is to research the appropriate methods to predict the spread trend of the existing topics on the internet,and to predict the future trend of the topics based on the historical trend information.The time series information of topic trend reflects the development law of topic trend itself,and contains abundant potential information.With the development of neural network,the time series prediction model based on LSTM,Transformer and BP neural network has been paid more and more attention.In this paper,we use Sina Weibo data to establish topic trend prediction model,with the emphasis on improving the neural network-based time series prediction model to meet the requirements of topic trend prediction.The main research work of this paper is as follows:1.In order to solve the problem of long-distance dependence and poor robustness of current hot Topic trend prediction models,this paper proposes a novel Topic prediction model TP-LS(Topic propagation base on LSTM Self-Attention),which combines LSTM and Self-Attention.First,the above information is fused based on the heat value,then,local temporal information and spatial structure information,global temporal and spatial structure information and feature extraction are coupled by LSTM network,Self-Attention mechanism and convolutional neural network,respectively,finally,the prediction results based on mathematical modeling method,complementary features and similar topics of statistical information are fed into the full link neural network to predict the heat trend values.The experimental results show that the mean square error,root mean square error and mean absolute error of the TP-LS model are reduced by 1%,4% and 5%,which indicates that the model is reasonable and effective.2.In view of the problem that the current topic trend prediction model is not effective in predicting explosive topics,the problem that the traditional method can not predict the explosive topic accurately by using the model of topic historical trend heat value.A new Topic prediction method TP(Topic prediction)is proposed.Firstly,a time series forecasting model A-GRU which fuses Arima and GRU is proposed to extract the historical trend information features of topics and combine the cascade information features as complementary features to enhance the forecasting ability of the model for explosive topics.The model is tested on Real Weibo data sets,and the results show that the model performs well not only when there is an explosive trend in topics,but also when there is a nonexplosive trend,the validity of the model is verified.