KGAT

Abstract

To provide more accurate, diverse, and explainable recommendation, it is\ncompulsory to go beyond modeling user-item interactions and take side\ninformation into account. Traditional methods like factorization machine (FM)\ncast it as a supervised learning problem, which assumes each interaction as an\nindependent instance with side information encoded. Due to the overlook of the\nrelations among instances or items (e.g., the director of a movie is also an\nactor of another movie), these methods are insufficient to distill the\ncollaborative signal from the collective behaviors of users. In this work, we\ninvestigate the utility of knowledge graph (KG), which breaks down the\nindependent interaction assumption by linking items with their attributes. We\nargue that in such a hybrid structure of KG and user-item graph, high-order\nrelations --- which connect two items with one or multiple linked attributes\n--- are an essential factor for successful recommendation. We propose a new\nmethod named Knowledge Graph Attention Network (KGAT) which explicitly models\nthe high-order connectivities in KG in an end-to-end fashion. It recursively\npropagates the embeddings from a node's neighbors (which can be users, items,\nor attributes) to refine the node's embedding, and employs an attention\nmechanism to discriminate the importance of the neighbors. Our KGAT is\nconceptually advantageous to existing KG-based recommendation methods, which\neither exploit high-order relations by extracting paths or implicitly modeling\nthem with regularization. Empirical results on three public benchmarks show\nthat KGAT significantly outperforms state-of-the-art methods like Neural FM and\nRippleNet. Further studies verify the efficacy of embedding propagation for\nhigh-order relation modeling and the interpretability benefits brought by the\nattention mechanism.\n

Keywords

Affiliated Institutions

• University of Science and Technology of China CN
• Shandong University CN
• National University of Singapore SG

Related Publications