Abstract
Recommender systems are one of the most effective technologies to help users filter the overload of information, and collaborative filtering (CF) is one of the most widely used techniques in recommender systems. However, CF algorithms have difficulty dealing with the problems such as the sparseness of data and the scalability for new users. As an alternative, an improved algorithm based on the preference (tendencies-based, TB) algorithm was proposed. In the proposed method: firstly, the user rating set was classified into different groups according to user rating preference and item rated tendencies; then the regression model was obtained by linear regression performed on each class. The improved model not only achieves higher accuracy in rating prediction on sparse datasets, but also greatly reduces computational complexity and space complexity. Through extensive experiments on three benchmark data sets, the results show that the improved approach increases recommendation accuracy by an average of 3.97% compared with TB algorithm.
Abstract
Recommender systems are one of the most effective technologies to help users filter the overload of information, and collaborative filtering (CF) is one of the most widely used techniques in recommender systems. However, CF algorithms have difficulty dealing with the problems such as the sparseness of data and the scalability for new users. As an alternative, an improved algorithm based on the preference (tendencies-based, TB) algorithm was proposed. In the proposed method: firstly, the user rating set was classified into different groups according to user rating preference and item rated tendencies; then the regression model was obtained by linear regression performed on each class. The improved model not only achieves higher accuracy in rating prediction on sparse datasets, but also greatly reduces computational complexity and space complexity. Through extensive experiments on three benchmark data sets, the results show that the improved approach increases recommendation accuracy by an average of 3.97% compared with TB algorithm.