Research On Trend Analysis And Prediction Of Network Hotspots

In recent years,natural language processing work related on social media information has attracted extensive attention,especially the monitoring and early warning of social emergencies and network sensitive information,analysis and prediction of the trend of emotional changes in public opinion.All of these mentioned above are of great research value.In this paper,We focuse on Sina microblog data.We analyze and calculate the emotional trends of hot topics,and make trend modeling based on historical microblog data to predict the future trends of hot topics.According to the characteristics of micro-blog data,a hot topic is divided into short-term and long-term topic.The trend of the two different topics were analyzed and predicted separately.We mainly study the characteristics of forecasting trends.The main research work of this paper is as follows:1.This paper proposes an emotion classification method based on joint deep learning,which divides microblog data into three categories: positive,negative,and middle,and extracts the features of text data by using CNN.The same dimension features are sent into LSTM to predict.The results show that the method is more accurate for sentiment classification of micro-blog data,and has a better effect on the emotional proportion of microblog's overall topic.2.We forecast trend analysis and trend of micro-blog short-term hot events,the method by calculating the data in the sample range,relevant indicators of influence events trend data values,2 hours is divided into 1 time periods were compared using different historical time data,to achieve the best effect in predicting the 2 time section.In the research of event trend prediction,the regression model is constructed according to the feature category to predict the topic heat.Comparison of the four kinds of prediction methods:uto regressive model,linear regression,GBDT and CNN,the experimental results show that the predicted 2 hours trend in the short term in the topic,the method of GBDT to achieve the best results based on when the prediction error is less than 5% for accurate prediction,accuracy of 79.1%.3.For the long term topic,topic temperature often changes with the development of the sub topic,so the prediction always lags behind.Based on the wave prediction on the basis of poor,the sub topic separation prediction method is put forward for training on the micro-blog data.The online LDA model is used to train microblog data on the same time slice to obtain the sub topic evolution and sub topic intensity.We divide the topic into 4 categories and use SVM to establish a classification model to predict the data between different peaks.The experimental results show that the accuracy of the proposed method is 86%,and the overall trend prediction has also achieved good results.