Research On The Detection And Evolution Analysis Method Of Bursty Topics Based On Social Media

With the rapid development of Internet technology,social media has become the main tool and channel for users to obtain information and communicate with each other.Microblog is one of the most popular social media at present,It has a huge number of online users,microblog information is published quickly,conveniently and in real time,and often becomes the first release site for bursty topics.Therefore,how to quickly and accurately detect bursty topics from the massive microblog data stream and analyze their evolution is of great significance.It can not only help users grasp the development trend of bursty topic,but also provide emergency measures for relevant departments,and provide theoretical support for public opinion analysis.This article takes microblog text stream as the research object,and aims to build a framework for microblog bursty topic detection and evolution analysis.In this framework,the related framework and algorithms for the bursty topic detection and evolution analysis are systematically studied.The main research work is as follows:(1)Problem analysis of bursty topic detection and evolution.First,the related concepts are introduced.Then,the challenges of current microblog bursty topic detection and evolution analysis methods are presented,and the microblog bursty topic detection and evolution analysis feamework is built.Finally,the characteristics of microblog data are analyzed and data preprocessing is introduced.(2)A microblog bursty topic detection method based on dynamic window is proposed.This method uses the word pair acceleration as the bursty feature,adaptively determines the range of the bursty topic window according to the velocity of the burst word pair in the microblog text stream,and the clustering algorithm based on improved nonnegative matrix factorization is used to obtain topic structure of microblog in the bursty topic window.The proposed detection method can not only reduce the time delayof bursty topic detection,but also improve precision and recall rate.and lay the foundation for the bursty topics evolution analysis.(3)A microblog bursty topic evolution analysis method based on extreme point is proposed.This method firstly divides the evolution time window by using the extreme point of the burst topic velocity,then obtains the topic cluster by clustering the microblog in the evolution time window,and finally obtains the evolution of bursty topic by calculating the similarity of the topics in the adjacent time window.The experimental results show that the method can describe the specific time of topic evolution in detail and improve the accuracy and recall rate of bursty topic evolution analysis.