GiGL: Large-Scale Graph Neural Networks at Snapchat (2502.15054v2)

Abstract: Recent advances in graph ML with the introduction of Graph Neural Networks (GNNs) have led to a widespread interest in applying these approaches to business applications at scale. GNNs enable differentiable end-to-end (E2E) learning of model parameters given graph structure which enables optimization towards popular node, edge (link) and graph-level tasks. While the research innovation in new GNN layers and training strategies has been rapid, industrial adoption and utility of GNNs has lagged considerably due to the unique scale challenges that large-scale graph ML problems create. In this work, we share our approach to training, inference, and utilization of GNNs at Snapchat. To this end, we present GiGL (Gigantic Graph Learning), an open-source library to enable large-scale distributed graph ML to the benefit of researchers, ML engineers, and practitioners. We use GiGL internally at Snapchat to manage the heavy lifting of GNN workflows, including graph data preprocessing from relational DBs, subgraph sampling, distributed training, inference, and orchestration. GiGL is designed to interface cleanly with open-source GNN modeling libraries prominent in academia like PyTorch Geometric (PyG), while handling scaling and productionization challenges that make it easier for internal practitioners to focus on modeling. GiGL is used in multiple production settings, and has powered over 35 launches across multiple business domains in the last 2 years in the contexts of friend recommendation, content recommendation and advertising. This work details high-level design and tools the library provides, scaling properties, case studies in diverse business settings with industry-scale graphs, and several key lessons learned in employing graph ML at scale on large social data. GiGL is open-sourced at https://github.com/Snapchat/GiGL.

An Overview of GiGL: Large-Scale Graph Neural Networks at Snapchat

The paper outlines the development and deployment of GiGL (Gigantic Graph Learning), a comprehensive library designed to facilitate the use of Graph Neural Networks (GNNs) at scale within industrial contexts, specifically at Snap Inc. The authors address the crucial challenge of scalability in deploying GNNs for large social graphs, such as those inherent to Snapchat's ecosystem, detailing both the technical design and business impacts realized over two years of utilization.

Scalability Challenges and the GiGL Solution

Industrial adoption of GNNs faces significant hurdles, primarily due to the massive scale of real-world graph data which often involves hundreds of millions of nodes and tens of billions of edges. GiGL is presented as a solution to these challenges, enabling efficient management of graph data and the execution of GNN workflows at this scale. It extends support for both supervised and unsupervised learning tasks such as node classification, link prediction, and representation learning. It integrates seamlessly with popular open-source libraries like PyTorch Geometric, simplifying integration for practitioners familiar with academic frameworks while addressing industrial requirements.

GiGL's Pipeline and Infrastructure

GiGL incorporates both tabularization and real-time subgraph sampling strategies for handling graph data, catering to varied industrial needs. The tabularization technique precomputes graph data for training and inference, thus enabling cost amortization and easy scaling across multiple tasks. This approach is well-suited for environments like Snap, where repeated graph training might be necessary for different product applications.

GiGL's pipeline includes a Data Preprocessor for transforming raw graph data, a Subgraph Sampler for subgraph generation, and a Trainer to facilitate model training. These components are orchestrated to support horizontal scaling across distributed systems, leveraging platforms like Kubeflow and VertexAI for efficient resource management. The authors also discuss a real-time sampling approach utilizing customized support from GraphLearn-for-PyTorch, providing adaptive graph access during training.

Industrial Applications and Impact

GiGL has been instrumental in various Snapchat applications, particularly in friend and content recommendation systems, wherein GNNs are used to improve the quality and diversity of recommendations. The paper details several successful deployments in friend recommendation, highlighting improvements gained through iterations on graph definitions, model architectures, and loss functions.

For instance, transitioning from a traditional graph-based retrieval system to one based on GNN embeddings has led to significant performance improvements across business metrics. Unique modeling techniques such as Stochastic EBR and supervised link prediction, adapted using user-defined labels, are covered, showcasing innovative uses of GNNs beyond conventional methodologies.

The implementation of heterogeneous graphs for content and advertisement recommendation is another noteworthy application of GiGL, where the complex interaction data between users and content are effectively modeled to drive engagement and conversions.

Future Directions and Community Contributions

The authors identify several areas for further exploration and potential improvement. These include advancements in link prediction techniques, such as subgraph GNN architectures, and the integration of LLMs to enhance node embeddings. Additionally, the paper emphasizes the development of cross-domain applications, which would leverage heterogeneous graphs to transfer knowledge across different types of relationships and interactions.

By open-sourcing GiGL, the authors aim to contribute to the broader graph ML community, encouraging further exploration of large-scale graph learning techniques and fostering collaboration in real-world applications of GNNs. The documentation and modular design of the platform ensure that it can be a valuable resource for researchers and practitioners interested in scalable GNN solutions.

Overall, the paper provides a detailed exposition of GiGL's capabilities and establishes its value in addressing the unique scale and complexity challenges of industrial GNN applications at Snapchat.