文摘
Collaborative filtering (CF)-based methods in recommender systems believe that the user’s preference of an item is the aggregation of the similar items or users. However, conventional item-based or user-based CF methods only consider either the item similarity or the user similarity. In this paper, we present hybrid-based methods for generating top-N recommendations in which both the item-item and user-user similarities are captured by the dot product of two low dimensional latent factor matrices. These matrices are learned using a stochastic gradient descent (SGD) algorithm to minimize two different loss functions, one is the squared error loss function and the other is the logistic loss function. A comprehensive set of experiments on multiple datasets is conducted to evaluate the performance of the proposed methods. The experimental results demonstrate the factored hybrid similarity methods (FHSM) achieve a superior recommendation quality in comparison with state-of-the-art methods.