AKUPP: attention-enhanced joint propagation of knowledge and user preference for recommendation systems

As knowledge graphs have attracted enormous attention from researchers, much effort has been invested in recommendation systems to mine user preferences effectively. In particular, knowledge graphs, which convey useful side information about users and items, can provide more accurate and explainable recommendations. When it comes to interactions between entities, however, the majority of existing work fails to incorporate high-order relations that ensure recommendation accuracy. This paper proposes attention-enhanced joint knowledge and user preference propagation (AKUPP), which integrates two types of knowledge propagation. The first is propagating user preferences based on the users' history of interacting items through ripple sets. The second propagation employs an attention mechanism to emphasize the important semantics of relations, and with multiple layers, high-order relations are explored. Therefore, we successfully incorporate both side information and high-order relations in the knowledge graph. We show, via extensive experimentation on real-world datasets, that our approach outperforms numerous state-of-the-art baselines in terms of performance and accuracy.

Automated summary

This paper introduces an attention-enhanced joint knowledge and user preference propagation method, which successfully incorporates both side information and high-order relations into the knowledge graph. Through extensive experimentation, the approach outperforms numerous state-of-the-art baselines in terms of performance and accuracy.