Research And Improvement Of Personalization On Collaborative Filtering Recommendation Algorithm

Personalized recommendation does not require the user to take the initiative to provide information,but by mining the user's historical behavior,access to the user's hobbies,habits,personality and other information,to further analyze the needs of users to determine the individual fonn of active recommendation to users' personal characteristics and interest preferences.The collaborative filtering recommendation algorithm has its own advantages,such as the type of recommendation object,the user feedback information and so on.However,there is no difference between the user's similarity calculation methods in the user-based collaborative filtering recommendation algorithm,and their own interest information and other issues.The use of a single user-based recommendation or item-based recommendation is not sufficient for the potential association between users and items,and will result in poor personalization of the recommended algorithm.In this paper,the above two issues as a research focus,the main research work is as follows:(1)For lacking the consideration of the user interest in the similarity calculation method of the traditional user-based collaborative filtering recommendation algorithm,which affects the personalized performance of the recommended algorithm,this paper proposes the UICF algorithm to improve the personalized recommendation performance of traditional algorithm through introducing the user-interest weight to improve the users'similarity calculate,and alleviates the pressure of above problems on the recommendation results by using the look-up table to clean up the data.The validity of the UICF algorithm is verified by comparison with the traditional user-based collaborative filtering recommendation algorithm and the existing algorithm.(2)Single-use user-based recommendation or project-based recommendation cannot be effectively utilized the potential link between user and item information,resulting in unsatisfactory results.This paper adopts the method of mixing these two algorithms for the use of user-based.Neither the commonly used general nor the average coefficients cannot be a good combination of user and project information,in the fusion algorithm using a linear coefficient of the way to protect the recommended performance of the hybrid algorithm.(3)If using the traditional user-based or item-based recommendation algorithm in the fusion algorithm,the recommended effect is still unsatisfactory.In this paper,the ITCF algorithm is proposed,and the dual weighting factor is adopted in the fusion algorithm.The UICF algorithm proposed in this paper is used in the user recommendation algorithm.The project time weight is introduced in the project similarity calculation to solve the problem of project timeliness and user interest the impact of migration.The new algorithm not only can accurately grasp the interests of users from the massive information,in the user or a project in the case of less information can also guarantee the recommended results.Finally,the experimental results are compared to prove the effectiveness of the algorithm.Finally experiment in the MovieLens data set.The validity of the UICF algorithm is verified by comparison with the traditional user-based cooperative filtering recommendation algorithm and the existing algorithm.By comparing with the traditional user-based collaborative filtering recommendation algorithm and using the mean-factor mixed algorithm,it is proved that the effectiveness of the ITCF algorithm.