Towards Practical Large-scale Dynamical Heterogeneous Graph Embedding: Cold-start Resilient Recommendation

TL;DR

This paper presents a scalable, two-stage framework combining a graph transformer and incremental embedding for cold-start resilient, real-time updates in large-scale dynamic heterogeneous graph applications, notably improving recommendation performance.

Contribution

It introduces HetSGFormer and ILLE, a novel combination of static and incremental learning methods for efficient, cold-start resilient dynamic graph embedding at scale.

Findings

Achieved up to 6.11% lift in Advertiser Value on billion-scale graphs.

ILLE module improved embedding refresh timeliness by 83.2%.

Framework is practical for production deployment of dynamic graph learning.

Abstract

Deploying dynamic heterogeneous graph embeddings in production faces key challenges of scalability, data freshness, and cold-start. This paper introduces a practical, two-stage solution that balances deep graph representation with low-latency incremental updates. Our framework combines HetSGFormer, a scalable graph transformer for static learning, with Incremental Locally Linear Embedding (ILLE), a lightweight, CPU-based algorithm for real-time updates. HetSGFormer captures global structure with linear scalability, while ILLE provides rapid, targeted updates to incorporate new data, thus avoiding costly full retraining. This dual approach is cold-start resilient, leveraging the graph to create meaningful embeddings from sparse data. On billion-scale graphs, A/B tests show HetSGFormer achieved up to a 6.11% lift in Advertiser Value over previous methods, while the ILLE module added another 3.22% lift and improved embedding refresh timeliness by 83.2%. Our work provides a validated framework for deploying dynamic graph learning in production environments.