Research On Topic Recognition And Heat Prediction Of Microblog Information Based On Multi-feature Fusion

In recent years,with the rapid development of Web 2.0,online social networks such as social media have become the leading role of Internet services and applications.As of December 31,2018,there were as many as 60 Weibo verticals,with monthly active users increasing by 70 million to 462 million for three consecutive years;Twitter had 126 million daily active users and 321 million monthly active users.A large number of active users observe and understand the world through social media,making hundreds of millions of information spread rapidly on the Internet every day.These online information records people’s behaviors of participating in,supervising and affecting the world,and provides valuable information resources for many academic researchers.This paper takes Sina Weibo as the main research scenario,aiming at the organization and ordering of Weibo information,and proposes the research of multi-feature fusion of microblog information topic recognition and popularity prediction.Specifically,it includes two aspects,namely the research of topic recognition combining the features of shallow text and deep semantic features,and the research of topic heat prediction based on self-motivated Hawkes process.In the topic recognition of Weibo information,it is considered that the short text of Weibo is often spoken heavily in information expression and there are many network terms,which leads to the lack of concentration of information expression.In this paper,lexical feature word embedding method is used to extract deep semantic features,which are combined with LDAbased text distribution features to eliminate data processing problems caused by the sparsity and irregularity of Weibo short text.Then,we use clustering algorithm to realize the topic recognition.Finally,a comparison test is performed on different text representation methods,such as LDA features,TF-IDF values,and word vectors.The experimental results show that the method using the fused features has achieved more topic recognition effects.In the prediction problem of topic heat of Sina Weibo,this paper firstly design the topic heat calculation formula based on the user characteristics and propagation characteristics of Weibo.Specifically,it includes user’s authentication information,tweets,followers,and Weibo retweets.Then,based on the early dissemination history of Weibo information,predict the future trend of Weibo topic and specific heat values.Specifically,according to the early time series information of each forwarding arrival of Weibo message,the self-motivated Hawkes process is used to model the dynamic process of information transmission,and then combined with the topic heat calculation formula to complete the final prediction research.Finally,the prediction results are compared with other classic prediction methods.The experimental results show that the model in this paper has better performance for topic heat prediction.