Research On Collaborative Filtering Recommendation Algorithm Based On Matrix Completion

The rapid development of Internet technology has brought a new era-the era of big data.People are generating and using large amounts of data anytime,anywhere.Massive data resources have brought great convenience to people’s lives,but it becomes very difficult for people to quickly find the information they want.This is the "information overload" problem.In the recommendation system,collaborative filtering technology is one of the most effective ways to solve the problem of information overload.Therefore,this paper takes the collaborative filtering recommendation algorithm as the research object,researches and analyzes the problems in its practical application,and proposes a new and improved algorithm based on the matrix filling algorithm.First of all,in view of the problem that the high missing rate of the original data affects the recommendation effect,a collaborative filtering algorithm based on matrix filling is proposed.Considering the sparseness and low rank of the user-item scoring matrix,the scoring matrix is filled by the OptSpace algorithm in the matrix filling technology,and then the new scoring matrix obtained after the filling is used,and then a collaborative filtering algorithm is used for recommendation.Secondly,the traditional collaborative filtering algorithm and matrix-filling-based collaborative filtering algorithm based on similarity to find the nearest neighbors.The search range is the entire data space,which is too computationally intensive and time-consuming,and the algorithm recommendation efficiency is not high.In addition,the k-means clustering method is hard partitioning of data objects,which does not meet the actual situation of the recommendation system,and the clustering effect is unstable.This paper proposes a collaborative filtering algorithm based on fuzzy clustering and matrix filling.Before performing collaborative filtering recommendation on the filled scoring matrix,users are clustered according to the user’s information.This algorithm reduces the search range of the nearest neighbor users to a certain cluster,reduces the calculation amount of similarity calculation,and improves the efficiency of the algorithm.Finally,in order to verify the effectiveness of the proposed algorithm,the validity of the algorithm is verified on the Movie Lens dataset in several experiments.The relevant parameters of the algorithm are determined through experiments to find the optimal number of clusters and the number of nearest neighbors.Then,the fuzzy clustering in the collaborative filtering algorithm based on fuzzy clustering and matrix filling is replaced by k-means clustering,and then the accuracy of the two algorithms is compared.Experiments show that the division of data objects by fuzzy clustering is more in line with actual conditions,The clustering effect is better.Then we compared several collaborative filtering recommendations according to the running time of the algorithm.The numerical experimental results show that the algorithm based on the clustering method improves the efficiency of the operation;in the case of determining the relevant parameters,a comparison test is performed on four different collaborative filtering recommendation algorithms.The recommendation algorithm proposed in this paper is superior to other recommendations Contrast algorithms.