Research On Collaborative Filtering Recommendation Algorithm Based On Improved Clustering And Matrix Factorization

With the rapid popularization of the Internet and the rapid expansion of e-commerce,the size of data in the world is rapidly increasing.People have entered the era of information overload from the era of insufficient information resources.In order to solve the problem of information redundancy caused by information overload,the personalized recommendation system is adapted to the needs of the times.It appears to alleviate the problems,but also improves the user's satisfaction with the recommended system services.Personalized recommendation system and recommendation algorithm has become a research hot spot and application difficulty.The traditional collaborative filtering recommendation algorithm usually finds the user's preference from the user's historical behavior data information of the project,so that the user is divided into groups according to the user's different preferences,and the target user is recommended with similar items.However,with the increasing number of users and items in e-commerce systems,data sparsity,low recommendation accuracy,etc.,have gradually become the key factors restricting the development of collaborative filtering recommendation algorithms.To solve the above problems,a collaborative filtering recommendation algorithm based on improved clustering and matrix factorization is proposed in this paper.The algorithm firstly uses matrix factorization to fill the original data with dimension reduction and its missing data.It also introduces a time decay function to preprocess the user's score,uses the user's interest vector to represent the user,and uses the project's attribute vector to represent the project.Based on this,the users and projects are respectively clustered by the k-means clustering algorithm;then use the improved similarity measure method to find the user's nearest neighbor and project recommendation candidate set in the cluster so as to generate a recommendation for the user.The experimental results manifest that the proposed algorithm not only can effectively handle the problem of cold start caused by sparse data and the appearance of new projects,but also can reflect the change of user's interest in multiple dimensions.The accuracy of the recommendation has also been improved.