A Hybrid Recommendation Method with Double SVD Reduction

Abstract

An issue related to recommendation is the requirement of considerable memory for calculating the recommendation score. We propose a hybrid information recommendation method using singular value decomposition (SVD) to reduce data size for calculation. This method combines two steps. First, the method reduces the number of documents on the basis of the users’ rating pattern by applying SVD based on collaborative filtering (CF). Second, it reduces the number of terms on the basis of the term frequency pattern of the reduced documents by applying SVD based on content-based filtering (CBF). The experimental results show that the proposed method has almost the same mean absolute error (MAE) as the SVD-based CBF. Originally, our data set has 9924 terms. The SVD-based CBF reduces the number of terms to 45 and the proposed method to 15 while preserving the same MAE. This means that the proposed method is effective for calculating recommendation.