Efficient Dynamic Graph Representation Learning at Scale

TL;DR

This paper introduces EDGE, an efficient algorithm for dynamic graph representation learning that scales to large datasets and achieves state-of-the-art performance by selectively modeling temporal dependencies.

Contribution

The paper presents a novel scalable algorithm, EDGE, that improves dynamic graph learning efficiency by selectively expressing temporal dependencies, enabling deployment on large industrial datasets.

Findings

Scales to graphs with millions of nodes and hundreds of millions of events

Achieves new state-of-the-art performance in dynamic graph tasks

Improves computational efficiency through selective temporal dependency modeling

Abstract

Dynamic graphs with ordered sequences of events between nodes are prevalent in real-world industrial applications such as e-commerce and social platforms. However, representation learning for dynamic graphs has posed great computational challenges due to the time and structure dependency and irregular nature of the data, preventing such models from being deployed to real-world applications. To tackle this challenge, we propose an efficient algorithm, Efficient Dynamic Graph lEarning (EDGE), which selectively expresses certain temporal dependency via training loss to improve the parallelism in computations. We show that EDGE can scale to dynamic graphs with millions of nodes and hundreds of millions of temporal events and achieve new state-of-the-art (SOTA) performance.