Research On Prediction Of Online Topics' Popularity Patterns

In the Internet era,online social platforms,as representatives of new media,have attracted a large number of users because they can efficiently and fast disseminate information,and have become important platforms for public opinion evolution and spreading.In recent years,there have been several social topics that originated from the social platforms first,then spread rapidly,and finally were noticed by other media and even society.Therefore,the spreading trend of a topic on online social platforms can be seen as an alarm of public opinion bursting.Monitoring and forecasting the spreading trend of public opinions in such media platforms is an important means of early warning and controlling Internet public opinion.At present,most existing work is to predict whether the content is popular or not,popularity level and popularity value,but the topic popularity pattern prediction still remains to be explored.Compared with the three above,popularity pattern prediction can provide more concrete topic spreading details for public opinion monitoring and early warning.This paper proposes a method which combines clustering and classification models to predict the popularity patterns of online topics.This method does not rely on the early time series data of topic propagation,so it can predict the future popularity pattern at the initial stage of topic publishing.First,K-SC,a time series clustering algorithm is used to obtain the basic types of topic popularity patterns on the social platform;then,multi-dimensional features related to topics are evaluated and sifted to build a topic feature space;based on the feature space and machine learning methods,we establish a topic popularity pattern prediction model and evaluate it.The experimental results show that the model can effectively predict the popularity patterns of new topics by inputting certain initial features,receiving an accuracy of 89.4%.At last,this paper explores the application of topic popularity patterns,and finds that patterns of popularity can help improve and guide the specific prediction of topic popularity volumes,and also help choose prediction methods and parameters.