An improved collaborative filtering method for recommendations' generation

Wujian Yang, Zebing Wang, Mingyu You

Abstract: Among the recommender system technologies, collaborative filtering system, which employs statistical techniques to find a set of customers who have a history of agreeing with the target user, has achieved widespread success on the e-commerce site. Although collaborative filtering system overcomes almost all the shortcomings of content-based systems, it is still reported having some limitations just like sparsity and scalability. In this paper, clustering using representatives algorithm is used to generate a new cluster-product matrix from original matrix. Based on the new matrix, traditional way is adopted to find the nearest neighbors. And at last a formula is given to generate the top-N recommendations. The experiment results suggest that the improved collaborative filtering method can increase the accuracy of the recommendations and the efficiency of the system.