Improving missing values imputation in collaborative filtering with user-preference genre and singular value decomposition

One of the major concerns in collaborative filtering is sensitive to data sparsity. The other word, missing values are occurred when the customers rate to a few products or services, which bring about to less accuracy of the recommendation. Although the centroid of cluster and SVD are able to solve Sparsity problem, their drawbacks are 1) imputed mean is not derived from user preference and 2) imputed mean does not reflect to the real distribution since imputed mean comes from the average. Therefore, we propose “SVDUPMedianCF” in order to solve the defect of the traditional approach which is an imputation missing value by filling the missing values for each customer with the cluster centroid, obtained from K-means algorithm, of such customer along with singular value decomposition (SVD) in collaborative filtering. According to the experimental evaluation based on MovieLens dataset by using 5-fold cross validation, it has found that imputing missing values with the proposed model presents the lowest mean absolute error when comparison with traditional approach significantly. From the experimental result, the proposed model can improve the quality of recommendation results with significant difference (p<;0.05).