Design And Implementation Of Personalized News Recommendation System Based On User Behavior

A large amount of news is produced on the Internet every day.In order to avoid people being drowned in the sea of information and solve the dilemma that people cannot obtain information efficiently,a news recommendation algorithm is used for recommendation,so that users can obtain information of interest,and then satisfy people’s needs.personalized news reading needs.A lot of new news will be generated on the news app,and outdated news will disappear,which will bring the cold start problem of items.Not only is there less explicit feedback from users on news apps,but also implicit user feedback is sparse.Users are sometimes interested in a certain type of news,and then the interest will change over time.The traditional recommendation based on user browsing behavior,due to the use of RNN(or LSTM,GRU,etc.),limits the parallel ability of the model.In addition,the information will be lost in the process of calculation,and the gate mechanism such as LSTM is still powerless for the particularly long-term dependence phenomenon.In order to solve the above problems,the main work is as follows:(1)Analyze the requirements of the personalized news recommendation system,divide the relevant functional modules,and implement a fully functional news system using Web-related technologies,mainly including the following functions:registration,login,personal center,channel management,list display of recommended articles,search articles,view articles,like,comment,favorite,follow and other common functions,and finally carry out the function test of the system.(2)This paper adopts mixed recommendation,which is based on content recommendation,Fastformer recommendation and popular recommendation.Among them,the Fastformer model can achieve good results in news recommendation.Content recommendation can recommend new or not very popular news and is highly interpretable.Popular recommendation by big data technology can solve the related cold start problem and data sparsity problem.Therefore,in the end,the advantages of the three recommendation methods are combined,the deficiencies are made up,and a hybrid recommendation algorithm is obtained,which makes the recommendation effect better.(3)Use big data technology to complete the calculation of popular news,use real-time processing and analysis technology to quickly control user browsing behavior and changes in interests,and conduct real-time analysis of the use of news apps by users in the system,including:click on news categories and other dimensions.,exposure,collection,etc.are calculated.Calculate PV,UV,number of bounces,number of pages entered(session count),duration of continuous access,etc.according to the dimensions of channels,regions,versions,and new and old users.