An improved matrix factorization based model for many-objective optimization recommendation

Abstract

As the application scenarios of recommendation algorithms are becoming increasingly complex, the efficiency of traditional recommendation algorithm based on accuracy is no longer satisfied. To solve this problem, an improved matrix factorization based model for many-objective optimization recommendation is proposed to simultaneously optimize the four recommendation objectives of novelty, diversity, accuracy, and recall. As a novel double-layer recommendation model, two improved algorithms are composed: 1) For the bottom layer, an improved matrix factorization algorithm with additional regularization constraints is used to predict unknown item ratings; 2) For the top layer, the recommendation list is optimized by a many-objective evolutionary algorithm. Comprehensive experiments demonstrate that the proposed model can effectively improve the four recommended evaluation metrics. And a recommended list with novel and diverse items is provided for users in a more efficient way while maintaining accuracy.