Scaling knowledge graph embedding models for link prediction

Abstract

Developing scalable solutions for training Graph Neural Networks (GNNs) for link prediction tasks is challenging due to the inherent data dependencies which entail high computational costs and a huge memory footprint. We propose a new method for scaling training of knowledge graph embedding models for link prediction to address these challenges. Towards this end, we propose the following algorithmic strategies: self-sufficient partitions, constraint-based negative sampling, and edge mini-batch training. The experimental evaluation shows that our scaling solution for GNN-based knowledge graph embedding models achieves a 16x speed up on benchmark datasets while maintaining a comparable model performance to non-distributed methods on standard metrics.