Hot Search List Of Microblog Topics And Hot Topics Prediction

As one of the most popular mobile social software,microblog has become the focus of social network public opinion.In today's society,attention is the wealth code.With a lot of attention,people can get all kinds of benefits from it,but it also gives criminals opportunities,such as spreading fake news,implementing network violence and so on.Therefore,many scholars have conducted in-depth research on the prediction methods of microblog topic popularity,and these studies have also brought help to the social public opinion supervision and network marketing on the microblog platform.However,almost all the researches on the prediction of microblog topic popularity ignore the important influence of microblog topic hot search list on topic popularity.In many cases,whether a topic can enter the hot search list has a crucial impact on its popularity development.The topic entering the hot search list can get a huge amount of exposure,which may make its future popularity continue to surge,but the topic that can not enter the hot search list,even if it has a certain degree of heat in the early stage,it is likely to decline rapidly.This paper focuses on microblog topic hot search list,which is an important mechanism in microblog.According to the very early data of the topic,it predicts whether the topic can enter the hot search list in the future,and on this basis,it predicts whether the topic entering the hot search list can become a hot topic.Firstly,this paper makes an in-depth study and analysis on the influencing factors of microblog topic popularity and the characteristics of microblog topic hot search list.Through a large number of statistical analysis,this paper found that the vast majority of topics contain generally 20-40 microblogs or users to participate in the discussion when entering in the hot search list,and some of these users are opinion leaders.Then,4804 topics,a total of 192160 microblogs collected from the microblog platform,are used for experimental analysis.The improved method based on multivariate complex time series feature extraction and data mining model is used to predict the topic popularity.The experimental results show that this method has higher prediction accuracy than the original method.Finally,this paper also analyzes the features that play an important role in topic popularity prediction.Followings are several innovations in this paper:First,previous research mainly focuses on predicting the specific heat of the topic at a certain time in the future,so that the accuracy rate will not be too high,and the significance of such research in real life is not very large.This paper studies the important mechanism of microblog hot search list ignored in previous research,and divides the topic heat into three levels: that is,the topics that can not log on the hot search list,The topics with general popularity are listed in the hot search list,and the topics with high popularity are listed in the hot search list.Then this paper will predict them.Secondly,in the existing research,prediction methods can be roughly divided into three kinds: similarity comparison method based on heat time series,discrete decomposition and data mining model method based on heat time series,and multivariate feature extraction and data mining model method based on influencing factors.The method used in this paper belongs to the third of them,and is improved on the basis of previous research.The previous literature mainly used multivariate simple description statistical feature extraction method.This method can not make the model fully learn the time series features of topic data.In this paper,the multiple complex time series feature extraction method,such as permutation entropy,approximate entropy,linear regression analysis,autoregressive coefficient,first-order difference average,can fully extract the nonlinear,periodic,volatility,unpredictability and other features of the topic data,so as to improve the prediction accuracy of the model.Thirdly,this paper also uses some new features that have not been used in previous studies,such as the user's microblog authentication type,microblog rating and sunshine credit rating,and whether the topic text contains keyword features such as "response","voice" and "voice".The addition of these features helps to extract more helpful information.