A New-User Cold-Starting Recommendation Algorithm Based on Normalization of Preference

Cold-starting problem of recommender system has attracted much attention. In the case of cold-starting, the extreme sparsity of ratings would induce poor performance of traditional recommendation algorithms. This paper presents a new algorithm to deal with the issue of cold-starting by taking the preference of user´s ratings into consideration. After normalizing historical rating matrix, two-stage weighted prediction with user similarity is proposed, then the predicted rating value can be obtained by inverse normalization. The experimental results indicate that the method can not only guarantee good recommendation performance in the condition of user cold-starting, but also keep the recommendation consistency when the rating matrix is in normal state.