Study On Hot Topic Recognition And Heat Trend Prediction Based On Short Text

In the information age with the rapid development of science and technology,people are more and more accustomed to using the Internet to conduct activities such as obtaining,discussing and disseminating information.Netizens continue to pay attention to all kinds of content on the Internet,which is full of massive amounts of data,and express their opinions and emotions on them.As a representative social network platform based on short texts,Weibo can generate data that reflects rich user characteristics.These data contain unlimited social and commercial value.Therefore,how to discover the real and repeated topics that netizens are paying attention to in a timely manner based on text characteristics and from the perspective of users,and to predict the trend of these topics has become the focus of relevant government departments and enterprises,and it is also important for the entire harmonious online public opinion environment.Construction,optimization of search engine services,improvement of personalized decision support,and perfection of information services all play a vital role in promoting.The purpose of this paper is to conduct research on the recognition and heat prediction of hot topics arising from online public opinion events for short texts of Weibo,that is,to achieve hot topic detection and heat tracking based on text features and user characteristics.The research content of this paper can be divided into the following three aspects:(1)Topic extraction based on short text.In this paper,topic extraction technology based on dynamic embedding model BERT and improved K-means clustering algorithm is used to extract topics from short texts of Weibo from the perspective of text features,and then through empirical analysis,the event is extracted from Internet public opinion events Number of topics generated and topic content.These topics reflect the angle and content that netizens really pay attention to the online public opinion event,and thus can accurately and timely obtain the netizen's attention on the online public opinion event.This has further achieved a better grasp of the changes in online public opinion events and subsequent correct guidance of public opinion.(2)Hot topic recognition based on short text.First,it introduces the difference between hot topics and topics and the significance of studying hot topic identification,and then quantitatively analyzes hot topics from the perspective of user characteristics and user behaviors,and proposes two degrees of concern that affect the changes in topic popularity:key opinion leader attention and attention based on user behavior.Finally,the topic heat value is calculated by the constructed heat formula and the hot topics are successfully identified.(3)Trend analysis of hot topics.In this paper,the LSTM model is used to predict the trend of hot topics based on the three dimensions of data likes,reposts,and comments on the identified hot topics.Through the training of the model and the verification of the effect,a predictive model with a better fit was successfully obtained.Combined with the empirical analysis,the prediction results are compared and analyzed according to the online public opinion life cycle theory and the constructed heat decay function,so as to conduct a specific in-depth analysis of the impact of the hot topic's heat trend change at different stages,and propose Different objects respond to changes in the popularity of hot topics at different stages,and at the same time have a deep understanding of the principle of information dissemination.