Research on Knowledge Graph-Based Recommender Systems

Abstract

A Knowledge Graph-based Recommendation System (KG-RS) employs a knowledge graph to represent data and generate precise recommendations for customers based on the given information. In this paper, we first investigate the various filtering techniques commonly utilized in recommendation systems and analyze the distinctions between Heterogeneous Information Networks (HINs) and Knowledge Graphs (KGs). Then, we classify models based on their embedding methods, loss functions, entity representations, and the integration of additional information. Also, we classified the extra models they used to facilitate research. Our research demonstrates that item information is consistently included in knowledge graphs, while user information is not. Additionally, KG-RSs are progressing by incorporating more advanced models into the recommendation process, rather than complicating the knowledge graph itself.