Social Media Oriented Hot Topic Extraction And Popularity Prediction Based On Sentiment Analysis

In the era of social networks,microblog,WeChat and forums,as the representatives of the social media,have become dominant platforms for users to express their opinions and acquire freshest information.However,it is so difficult for users to dig out the hot topics and monitor their trends timely and effectively in such a huge social network with large amounts of unstructured data.The hot topic extraction and popularity prediction studied in this thesis is aimed at solving this problem.If the hot topics in social networks can be monitored and analyzed effectively,it will be of great significance for individuals to obtain the hotspots and-help to make decision for companies and governments.Hot events and topics discussed in social networks will trigger netizens'emotional fluctuation,leading to a sharp increase in the number of emotional microblogs.Thus in view of the fact that the occurrence of emergencies is often strongly associated with public emotion,this thesis designs a topic extraction model based on sentiment analysis.Firstly data acquisition is conducted.Then after a series of data processing,a self-supervised sentiment classification method is used to classify microblogs.Next,Single-Pass algorithm is improved for topic clustering to get topic clusters and the results are represented by word clouds.Experimental results then prove the feasibility of the proposed approach and its good performance on topic extraction.Hot topics in social networks are influenced by a variety of factors.And the time series of topic popularity is not simply linear,but time-varying,randomly mutated and nonlinear.Therefore,this thesis introduces the LSTM networks which is seldom applied in popularity prediction but has a good effect on nonlinear time series prediction.On the basis of the LSTM model,this thesis proposes an AB-LSTM model.The model applies ensemble learning methods to enhance the performance of popularity prediction.The results of thorough experiments demonstrate its effectiveness and superiority.