Knowledge Graph-Based Multi-Objective Optimization Recommendation Model

Abstract

The current recommendation application scenario is more complex, and the traditional recommendation algorithm can no longer meet the needs of user diversity. To solve this problem, this paper proposes an improved multi-objective optimization recommendation model based on knowledge graph to optimize four recommendation goals of novelty, diversity, accuracy and recall simultaneously. A user preference set is constructed based on user behavioral preferences and item-related characteristics before recommendation, which provides the basis for subsequent recommendations. Two improved algorithms are proposed in this paper: 1) at the bottom layer, a knowledge graph embedding algorithm with variable weighted scoring function is used to transform the association information between items, into the relationship between vectors; 2) for the top layer, a multi-objective evolutionary algorithm is used to optimize the recommendation list. Comprehensive experiments show that the model can effectively improve the evaluation metrics of the four recommendations. And it provides users with a recommendation list of items containing novel and diverse items in a more efficient way while maintaining accuracy.