The Research Of Collaborative Filtering Based On Improved K-means And Singular Value Decomposition

Collaborative Filtering is a common method to reduce the information overload, but most of Collaborative Filtering Algorithms for the current prevailing have low scalability and data sparseness problem. Aiming at these problems, a comprehensive optimization program is proposed for data sparsity and the scalability of recommendation algorithm on the basis of traditional SVD-based Collaborative Filtering algorithm and Collaborative Filtering algorithm based on K-means. This paper mainly includes the following 3 aspects:Firstly, The Collaborative Filtering method based on K-means, although the search target user narrowed it down to the nearest cluster, greatly reducing the amount of computation, and improved the real-time and scalability. But when users’ rating data is high-dimensional sparse, the recommendation accuracy of Collaborative Filtering Algorithms based on K-means tends to be lower. Aiming at this problem, this paper uses the SVD dimension prediction fill technology to fill the non score of the original basic data, improve the anti sparsity of recommendation algorithm.Secondly, Collaborative filtering algorithm based on SVD although overcome data sparseness problem, but Collaborative Filtering algorithm needs to search for the target user’s nearest neighbors in the whole data space, often very large amount of computation and low scalability. Furthermore, in order to avoid the slow convergence rate caused by the random selection of the initial clustering centers of K-means clustering and leads to unstable clustering results, and there is local minima problem. In this paper, we use the improved K-means clustering method to search for the nearest neighbors of target user narrowed it down to the nearest cluster, greatly reducing the amount of computation, improve the scalability of the algorithm.Thirdly, In the case of user rates of high-dimensional sparse data, This paper presents a Collaborative Filtering Recommendation based on improved K-means algorithm and Singular Value Decomposition. First, filling the missing ratings by SVD, and then implement improved K-means Clustering Algorithm in the filled matrix. Finally, implement Collaborative Filtering Recommendation in the target clustering.In order to verify effectiveness of the proposed recommendation algorithm in this paper, the experiments are carried out on the MovieLens dataset. Experimental results show that compared with the SVD-based Collaborative Filtering Algorithm and Collaborative Filtering Algorithm based on K-means, it is recommended that the proposed algorithm has higher rating precision and quality of recommendation.