Graph Convolutional Neural Networks for Web-Scale Recommender Systems

  • 2018-06-06 01:26:33
  • Rex Ying, Ruining He, Kaifeng Chen, Pong Eksombatchai, William L. Hamilton, Jure Leskovec
  • 37


Recent advancements in deep neural networks for graph-structured data haveled to state-of-the-art performance on recommender system benchmarks. However,making these methods practical and scalable to web-scale recommendation taskswith billions of items and hundreds of millions of users remains a challenge.Here we describe a large-scale deep recommendation engine that we developed anddeployed at Pinterest. We develop a data-efficient Graph Convolutional Network(GCN) algorithm PinSage, which combines efficient random walks and graphconvolutions to generate embeddings of nodes (i.e., items) that incorporateboth graph structure as well as node feature information. Compared to prior GCNapproaches, we develop a novel method based on highly efficient random walks tostructure the convolutions and design a novel training strategy that relies onharder-and-harder training examples to improve robustness and convergence ofthe model. We also develop an efficient MapReduce model inference algorithm togenerate embeddings using a trained model. We deploy PinSage at Pinterest andtrain it on 7.5 billion examples on a graph with 3 billion nodes representingpins and boards, and 18 billion edges. According to offline metrics, userstudies and A/B tests, PinSage generates higher-quality recommendations thancomparable deep learning and graph-based alternatives. To our knowledge, thisis the largest application of deep graph embeddings to date and paves the wayfor a new generation of web-scale recommender systems based on graphconvolutional architectures.


Introduction (beta)



Conclusion (beta)