The Research On Collaborative Filtering Recommendation Algorithm And Its Application

The development of the global village makes the existing language become the main obstacle for the communication between people in the world.Based on this,the "little dewdrop" team of Northwest university proposed a "harmony" language that crosses languages,cultures,man-machine,time and space,aiming to provide a unified communication language for users all over the world.As the "harmony" language spread,the number of new users and ICONS increased dramatically.It is difficult for users to find the icons that they are interested in.So,it is necessary to introduce the recommendation algorithm into the domain of icons.Collaborative filtering algorithm is the most successful method to solve information overload in the recommendation algorithm.The main problems are data sparsity and cold start.Therefore,this dissertation studies,improves and applies the collaborative filtering algorithm to solve the above problems.The main work of this thesis are as follows:Firstly,the collaborative filtering algorithm based on user characteristic and time weight(UCTWCF)is studied.The user registration information is used to calculate the similarity of user characteristics.The time weight function including interest invariant time window T is introduced to reflect the variability of user interest.Get recommendations that better match the user's current needs.Experimental results show that this algorithm can solve the cold start-up problem of new users to a certain extent.The influence of data sparsity on recommendation results is reduced and recommendation quality is effectively improved.Then,the collaborative filtering algorithm based on trust relationship(TRCF)is studied..Trust network is built by trust transfer and broaden the user base for recommendation.The problem of low accuracy of similarity calculation caused by sparse data is alleviated.The algorithm makes recommendations by calculating the user's direct trust,indirect trust,comprehensive trust and weighted trust respectively.Experimental results show that the algorithm has higher accuracy of prediction score.At last,the icon recommendation system is designed and implemented.The system uses UCTWCF and TRCF,personalized icon recommendations can be provided to the user.Improved user satisfaction and promotes the promotion and use of "harmony" language.