Research And Application Of Collaborative Filtering Algorithm Based On Trust Clustering

With the rapid development of Internet technology,various types of data are growing rapidly,but too much information has brought too many choices to people.In order to solve the problem of information overload,the recommendation system emerges as the times require.Collaborative filtering is a relatively successful technology in the recommendation system,but it also faces challenges such as data sparsity and scalability.In order to solve the shortcomings of traditional algorithms and improve the user-based collaborative filtering algorithm,this paper analyzes and studies the high-dimensional matrix dimension reduction and improved user similarity calculation methods from two angles,and establishes a two-stage clustering recommendation Model that integrates trust factors.First,the sparse original user score data is preprocessed.The paper firstly predicts the user's score on the unrated item according to the similarity between the items,so as to fill the matrix vacancy value and make it become a dense matrix that can perform SVD dimensionality reduction.Then,the low-dimensional user implicit feature space is obtained by the SVD method,thereby improving the calculation efficiency.Then,a two-stage clustering collaborative filtering recommendation algorithm model incorporating trust factors is constructed.After obtaining the implicit feature space matrix of the user,the fuzzy clustering method is used to initially determine the neighbor user cluster of each user;then the trust transfer factor is added to the local trust metric,and the potential association between users is mined,and the problem of insufficient information available is alleviated.At the same time,the degree of trust of the user in the entire network is considered,and the global trust of the user is calculated.And combine the two to get a measure of comprehensive trust.Through the two-stage neighbor search,the accuracy of the neighbor query is improved.Finally,based on the project score of the user in the trusted neighbor group,the unrated item of the target user is predicted,thereby improving the accuracy of the recommendation.Finally,this paper design experiment relies on classical data sets.Firstly,the optimal parameters of the algorithm are found through experiments,and then substituted into the improved algorithm.When the number of neighbors takes different values,the MAE of three classical collaborative filtering algorithms and improved algorithms are calculated.By comparison,when the number of neighbors exceeds 20,the accuracy of the improved algorithm is higher than that of the other three algorithms.Then the improved algorithm is applied to the established movie recommendation system to realize the recommended function of the system,and the feasibility of the algorithm is verified.