Sparsity-Tolerated Algorithm with Missing Value Recovering in User-based Collaborative Filtering Recommendation ⋆

Personalized recommendation plays an important role in both e-commerce area and information filtering area. The neighborhood based collaborative filtering algorithm has already been used successfully. However, with the overwhelming explosion of Internet content, the problem of data sparsity has become more and more severe. The effect of data sparsity problem lies in both similarity computation and prediction generation, but very few works focus on the latter. This paper presents a hierarchy k-nearest neighbor collaborative filtering algorithm. It fills in the missing value by constructing multiple layers of nearest neighbors for users to generate better prediction. Experiments validated that the algorithm proposed in this paper achieved higher prediction accuracy with extreme sparse data.