Reproduce “Graph Convolutional Neural Networks for Web-Scale Recommender Systems” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Rex Ying, Ruining He, Kaifeng Chen, Pong Eksombatchai, William L. Hamilton, Jure Leskovec

PinSage introduces a highly scalable Graph Convolutional Network (GCN) algorithm, developed by Pinterest and Stanford University, for web-scale recommender systems, demonstrating a consistent 10-30% improvement in repin rate in production at Pinterest. This work enables deep graph embeddings to operate on a graph 10,000 times larger than typical GCN applications.

Key insights

  • Localized graph convolutions with importance-based neighborhood sampling and importance pooling are crucial for scaling GCNs to web-scale graphs.
  • The combination of hard negative sampling and curriculum training substantially improves the quality and robustness of learned graph embeddings.
  • System-level innovations like a producer-consumer minibatch pipeline and a distributed MapReduce inference framework are essential for practical deployment of GCNs at production scale.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.