Entity-driven user intent inference for knowledge graph-based recommendation

It has been proven that a knowledge graph (KG) has the ability to improve the accuracy of recommendations, owing to its capability of storing the auxiliary information of items in a heterogeneous structure. Recently, intent inference methods have been developed to explore the user preference information from a KG and user-item interactions and to help improve the recommendation accuracy. The inferred user intent can also be regarded as a part of the reason why the recommendation model recommends a certain item to a user. It is known that there are two types of information in a KG: entities and relations. However, existing recommendation models infer user intents from only the information contained in the relations in a KG. In this paper, we propose a new recommendation model, the entity-driven knowledge intent network (EKIN) to infer user intents using information from both entities and relations and make recommendations for users. For EKIN, we propose to construct an entity-driven user intent graph (EUIG) for each user. The EUIG exploits two types of information in a KG to infer user intents. A graph neural network is constructed with multi-hop propagation in the KG and EUIG to learn the representation of entities, relations and user intents. Moreover, we distill information on users' interactions based on their inferred intents and aggregate the interactions to encode the user characteristics. The experimental results on three real-world datasets demonstrate that the proposed EKIN outperforms the state-of-the-art KG-based recommendation models.