Knowledge Graph Convolutional Networks for Recommender Systems

Abstract

To alleviate sparsity and cold start problem of collaborative filtering based\nrecommender systems, researchers and engineers usually collect attributes of\nusers and items, and design delicate algorithms to exploit these additional\ninformation. In general, the attributes are not isolated but connected with\neach other, which forms a knowledge graph (KG). In this paper, we propose\nKnowledge Graph Convolutional Networks (KGCN), an end-to-end framework that\ncaptures inter-item relatedness effectively by mining their associated\nattributes on the KG. To automatically discover both high-order structure\ninformation and semantic information of the KG, we sample from the neighbors\nfor each entity in the KG as their receptive field, then combine neighborhood\ninformation with bias when calculating the representation of a given entity.\nThe receptive field can be extended to multiple hops away to model high-order\nproximity information and capture users' potential long-distance interests.\nMoreover, we implement the proposed KGCN in a minibatch fashion, which enables\nour model to operate on large datasets and KGs. We apply the proposed model to\nthree datasets about movie, book, and music recommendation, and experiment\nresults demonstrate that our approach outperforms strong recommender baselines.\n