An Improved Collaborative Filtering Algorithm based on Dimension Reduction and Improved Clustering

Abstract

Aiming at the problems of sparse rating data, complex similarity calculation and low recommendation accuracy in traditional collaborative filtering recommendation algorithm when processing large-scale data, this paper proposes a new improved collaborative filtering recommendation algorithm. Based on the traditional collaborative filtering algorithm, this algorithm first uses the PCA algorithm to reduce the dimension of the sparse user-item rating matrix; secondly, the dimension-reduced rating matrix is combined with user attributes to construct a matrix containing both ratings and user attributes, according to the matrix, bisecting K-means clustering is performed from the two dimensions of user and item respectively; then, collaborative filtering recommendation is performed from the two dimensions of user and item respectively; finally, weighted integration of the predicted score results of the two dimensions of the user and the item obtains the final predicted score. Experiments using MovieLens100K dataset show that the algorithm proposed in this paper improves the quality of system recommendations.