Graph Neural Network Combined Knowledge Graph for Recommendation System

摘要

With a view to increase recommendation systems accuracy and practical applicability, using traditional methods which are namely interaction model between users and items, collaborative filtering and matrix factorization cannot achieve the supposed results. In fact, the properties between users or items always remains as social and knowledge relations. In this paper, we have proposed a new graph deep learning model associated with knowledge graph with the aim of modeling the latent feature of user and item. We exploit the relations of items based on knowledge graph as well as the relationships between users in social. Our model supplies the principle of organizing interactions as a graph, combines information from social network and all kind of relations in the heterogeneous knowledge graph. The model is evaluated on real world datasets to demonstrate this method's effectiveness.