Research On Personalized News Recommendation Based On Partition Model

With the development of Internet technology,online browsing news has become an important means for people to obtain information.However,the explosive growth of network data makes users fall into the embarrassing situation facing of a lot of spam.How to quickly and accurately obtain the information the users need has been solved urgently.News recommendation,as an effective means to solve the problem,can find the interest of users in the news by analyzing the news content and user historical behavior data.Currently,news recommendation system commonly used technology has included with content-based recommendation technology,collaborative filtering technology and mixed recommendation technology.This paper focuses on the personalized news recommendation,for the problem that the user interest model is too rough and the inaccuracy of similarity measurement method in collaborative filtering technology.the following research work is done:1.The traditional user preference model is based on the user's historical behavior data,this modeling method is rather rough,ignoring the impact of different time browsing news on the overall preferences of users.In this paper,by dividing the time slice,the historical behavior data of different time periods are respectively modeled,and a time factor is given to each preference model,and finally the user's long-term interest preference model is obtained.2.In order to accurately describe the volatility of the user's current preference,this paper proposes to dynamically model the behavior data of the user's current time period,timely find out the types of news the user mainly concerned about currently,and make an accurate recommendation.3.In order to solve the problem of inaccurate calculation of similar users in collaborative filtering technology,in this paper,we propose a new similarity measure method named preference-behavior similarity measure,to calculate the similarity of users from two aspects:user's preference and behavior,and then weighted summation.Increased the accuracy of calculating similar users.4.Based on the long-term preference model,dynamic preference model and improved collaborative filtering technology,in this paper,it is recommended for users to meet the needs of users for different types of news.Finally,the paper uses the data set which provided from the platform of DataCastle large data competition for experimental,and have the content-based news recommendation method,the collaborative filtering recommendation method and the mixed news recommendation method as contrast models.The experimental results show that the accuracy,recall rate and F value are significantly improved in the news recommendation method presented in this paper,and has improved the user experience satisfaction.