A Collaborative Filtering Algorithm Based on Variance Analysis of Attributes-Value Preference

Collaborative filtering is the state-of-the-art and widely applied method in personalized recommendation systems. However, the problem of precision resulting from sparsity exists chronically. To address the issue, we develop collaborative filtering algorithm that incorporates the variance analysis of attributes-value preference, which can improve recommending precision further. What we operate on is based on the new user-item rating matrix that has been reduced in dimensionality via singular value decomposition. Firstly, user ratings can be mapped to relevant item attributes for establishing attributes-value preference (AP) matrix. Variance matrix of AP (VAP) is proposed to compute the similarity between users that incorporate with the mean of it. Thus, the rating prediction is calculated to generate the top-N items for target user. The experiment suggests that it can increase the precision of collaborative filtering recommendation.