Abstract:Recommender algorithms combining knowledge graph and graph convolutional network are becoming more and more popular recently. Specifically, attributes describing the items to be recommended are often used as additional information. These attributes along with items are highly interconnected, intrinsically forming a Knowledge Graph (KG). These algorithms use KGs as an auxiliary data source to alleviate the negative impact of data sparsity. However, these graph convolutional network based algorithms do not distinguish the importance of different neighbors of entities in the KG, and according to Pareto’s principle, the important neighbors only account for a small proportion. These traditional algorithms can not fully mine the useful information in the KG. To fully release the power of KGs for building recommender systems, we propose in this article KRAN, a Knowledge Refining Attention Network, which can subtly capture the characteristics of the KG and thus boost recommendation performance. We first introduce a traditional attention mechanism into the KG processing, making the knowledge extraction more targeted, and then propose a refining mechanism to improve the traditional attention mechanism to extract the knowledge in the KG more effectively. More precisely, KRAN is designed to use our proposed knowledge-refining attention mechanism to aggregate and obtain the representations of the entities (both attributes and items) in the KG. Our knowledge-refining attention mechanism first measures the relevance between an entity and it’s neighbors in the KG by attention coefficients, and then further refines the attention coefficients using a “richer-get-richer” principle, in order to focus on highly relevant neighbors while eliminating less relevant neighbors for noise reduction. In addition, for the item cold start problem, we propose KRAN-CD, a variant of KRAN, which further incorporates pre-trained KG embeddings to handle cold start items. Experiments show that KRAN and KRAN-CD consistently outperform state-of-the-art baselines across different settings.

An End-to-End Neighborhood-based Interaction Model for Knowledge-enhanced Recommendation

An Efficient Neighborhood-based Interaction Model for Recommendation on Heterogeneous Graph.

Exploring Multi-dimension User-Item Interactions with Attentional Knowledge Graph Neural Networks for Recommendation

Knowledge-aware fine-grained attention networks with refined knowledge graph embedding for personalized recommendation

Neighbor-Augmented Knowledge Graph Attention Network for Recommendation

Knowledge-Enhanced Recommendation with User-Centric Subgraph Network

KGCNA: Knowledge Graph Collaborative Neighbor Awareness Network for Recommendation

KGIE: Knowledge graph convolutional network for recommender system with interactive embedding

Feature interactive graph neural network for KG-based recommendation

BKGNN-TI: A Bilinear Knowledge-Aware Graph Neural Network Fusing Text Information for Recommendation

Knowledge Enhanced Graph Neural Networks for Explainable Recommendation

Knowledge Graph Enhanced Neural Collaborative Filtering with Residual Recurrent Network

Graph Enhanced Neural Interaction Model for Recommendation

Attention-Enhanced and Knowledge-Fused Dual Item Representations Network for Recommendation

Knowledge-aware Coupled Graph Neural Network for Social Recommendation

Learning Intents behind Interactions with Knowledge Graph for Recommendation

Knowledge graph fine-grained network with attribute transfer for recommendation

Knowledge Graph Double Interaction Graph Neural Network for Recommendation Algorithm

Attentive Knowledge-aware Graph Convolutional Networks with Collaborative Guidance for Personalized Recommendation

Kgat: Knowledge Graph Attention Network For Recommendation

KRAN: Knowledge Refining Attention Network for Recommendation