Research And Design Of Personalized News Recommendation System

The personalized news recommendation system is a recommendation tool.Users login the system,and the system can recommend interesting news for them on account of their historical behaviors.The personalized news collaborative filtering recommendation algorithm calculates the similarity of news based on user's historical behavior to complete similar news recommendations.This similarity calculation method cannot dig the characteristics of news,and exists sparse data problems.Meanwhile,the collaborative filtering algorithm does not take user's changing interests in time into account.To solve the problem of the sparse data in similarity calculation,this paper emphasizes the study on the calculation of text similarity at home and abroad,and puts forward the news similarity calculation method which is suitable for the news characteristics.Based on existing text similarity calculation method,the improved algorithm can comprehensively consider two characteristics which are the different importance of words in different parts,and the more important state of news headlines.Then combining the calculating of users' behavior similarity,we apply the improved news similarity calculation algorithm to the news recommendation system.To solve the problem of users' time-changing interests,this paper has studied the existing personalized news recommendation algorithm at home and abroad,and proposes a personalized news recommendation algorithm which adapts to the changing interest of users.Users' recent browsing news usually make greater contributions to users interests model.But users' behaviors can be repetitive which means the earlier interests may also have an impact on users' choices.Therefore,we establish users' recent interest model and interests' behavior model.Based on cooperative filtering algorithm,we obtain users ' stable interests model,and apply it to the recommendation algorithm.We use DataCastle's new network reading record as the experimental data set,and evaluate results through F-measure value and average absolute error.Compared with the similarity calculation recommended algorithms,the result shows the improved algorithm based on the similarity calculation method for news' characteristics has a more accurate calculate news similarity,because the F-measure value is 10.5% higher than other algorithms at the most.It means the improved algorithm effectively avoids data sparse problem.Compared with the traditional cooperative filtering algorithm and the existing recommended algorithm,the F-measure value of the improved algorithm is 11.5% higher,and the average absolute error decrease 8% at most,which means that the improved algorithm can reflect users ' interest better.At last,this paper completed the design and realization of the personalized news recommendation system.By analyzing demand analysis and general design,and applying the improved algorithm to the system,we finally implemented the recommendation system.