News Popularity Prediction Based on Jump Relationship and User Preference
The advent of the Internet era has accelerated the frequency of news updating and expanded the scope of news dissemination.It is dificult to mine and analyze the potential risk of news content only relying on manual work.Therefore,it is of great significance to build a model to predict the future popularity of news for a period of time to quickly control bad information and strengthen Internet information governance.In order to better predict the news popularity,this paper focuses on min-ing the relationship between news,and proposes a news popularity prediction method(that integrates the news jump relationship and user preference under the premise of moderately integrating user preference information).This method first combines the news content and the historical jump probability to generate the news jump relation-ship network,and uses the multi-task graph convolution matrix completion model MGCMC(multitask graph convolution matrix completion)proposed in this paper to predict the sparsely distributed and unbalanced jump probability matrix,so as to obtain the characteristics of the future news jump relationship network.When the news platform recommends a group of news in the state of dissemination to users,this method combines users'personalized preferences to predict their click behavior,and finally gains news popularity.The experimental results based on the real user-news interaction dataset Mind show that MGCMC performs better than the existing matrix completion and unbalanced prediction models,and the accuracy of user-news click prediction is higher,and the discovery of popular news is more accurate.
NETL
NSTL
万方数据
