Link Prediction with Graph Neural Networks in Production

WHAT IS IT?

Graph Neural Networks (GNNs) have emerged as the most effective approach for machine learning on graphs. Graphs are useful for describing most real-world applications (including financial graphs, social networks, ecommerce, and online marketplaces), where phenomena can be broken down into entities, their relationships, and their dynamic interactions. 

Link prediction is the task of predicting an edge between any pair of nodes - thus defining their relationships or interactions. Link prediction has a broad range applications including recommendations (predicting a user’s affinity to different products, content, ads, or other users which would lead to a user buying specific items, viewing specific content, or clicking on an ad in any ecommerce or online marketplace transaction networks), customer outreach (ranking top outreach strategies by channel, content, timing, and volume), drug discovery (predicting molecular interactions and activations while simultaneously predicting side effects based on biological structure), transportation (predicting traffic flows to optimize routes), and more.

This webinar will deep dive on techniques and best practices for building link prediction for production use cases, including recommendation and ranking systems. We’ll walk through the different approaches of link prediction using GNNs, walking through the typical flow of setting up the problem and executing the prediction, and we’ll show you how to put it together yourself with PyTorch Geometric. We’ll show you what the out-of-the-box experience looks like using Kumo’s state-of-the-art GNN platform in production.