GNNs scaled for friend recommendations with multi-hash embeddings and temporal sampling

Researchers have developed a scalable Graph Neural Network (GNN) system for friend recommendation on large social graphs, addressing challenges in production environments. The system utilizes multi-hash ID embeddings to significantly reduce embedding table size by over 98% while maintaining ranking quality. Additionally, it implements temporal neighbor sampling with timestamp-sorted CSR storage and binary search, drastically cutting down sampling costs. An online A/B test demonstrated a 16% increase in friend additions from recommendations and an 11.5% rise in unique friend adders compared to existing systems. AI

IMPACT This research offers a scalable solution for friend recommendation systems, potentially improving user engagement on social platforms.

RANK_REASON The cluster contains a research paper detailing a new method for scaling graph neural networks for friend recommendation.

Read on arXiv cs.IR (Information Retrieval) →

• A/B testing
• Calibrated Size Ratio
• Friend Recommendation System using Image Encryption and Deep learning
• graph neural networks
• information retrieval
• Multi-Hash User Embeddings
• Temporal Neighbor Sampling
• alphaXiv
• CatalyzeX Code Finder for Papers
• DagsHub
• Gotit.pub
• Hugging Face
• Influence Flower
• ScienceCast

• paper
• infra
• product

AI-generated summary · Google Gemini · from 2 sources. How we write summaries →