RankGraph-2: Lifecycle Co-Design for Billion-Node Graph Learning in Recommendation

2026-06-16

AI Summary

This work addresses the efficiency and effectiveness bottlenecks in billion-scale recommender systems arising from the disjointed design of graph construction, representation learning, and real-time serving. To bridge this gap, the authors propose a lifecycle-coordinated framework that jointly optimizes all three stages for the first time. Key innovations include a popularity-bias-corrected edge sampling strategy, an hourly-updatable self-contained graph structure, precomputed personalized PageRank for multi-hop neighborhoods, and a residual quantization clustering index with end-to-end co-training. The approach enables efficient U2U2I/U2I2I similarity retrieval, reducing serving costs by 83% in production while achieving 3.8× and 2.1× higher recall than GAT+DGI and PyTorch-BigGraph, respectively. It also improves CTR and CVR by 0.96% and 2.75%, and has stably supported over 20 online retrieval iterations.

Abstract

Graph-based retrieval at billion-node scale requires jointly solving three tightly coupled problems -- graph construction, representation learning, and real-time serving -- yet existing work addresses each in isolation. We present RankGraph-2, a framework deployed at Meta that co-designs all three lifecycle stages for similarity-based retrieval (U2U2I and U2I2I), where each stage's requirements shape the others. Serving requires a co-learned cluster index to avoid expensive online KNN -- this pushes index co-training into the training objective. Training benefits from the observation that similarity-based retrieval tolerates pre-computed neighborhoods, eliminating online graph infrastructure -- this requires construction to produce self-contained data. Construction must also support hour-level refresh for item coverage. Acting on these cascading requirements, RankGraph-2 reduces hundreds of trillions of edges to hundreds of billions via subsampling with popularity bias correction, pre-computes multi-hop neighborhoods via personalized PageRank, and co-learns a residual-quantization cluster index that reduces serving computational cost by 83%. This lifecycle co-design enables a simple architecture to achieve 3.8 x higher recall than a GAT + Deep Graph Infomax model on a bipartite graph and 2.1 x higher than PyTorch-BigGraph on item retrieval. RankGraph-2 delivers up to +0.96% CTR and +2.75% CVR, and has powered 20+ retrieval launches across major surfaces.

Problem

• billion-node graph
• graph-based retrieval
• lifecycle co-design
• similarity-based retrieval
• real-time serving

Innovation

• lifecycle co-design
• billion-node graph learning
• co-learned cluster index
• pre-computed neighborhoods
• popularity bias correction