Few-shot graph link prediction with domain adaptation

Hao Zhu, Mahashweta Das, Mangesh Bendre, Fei Wang, Hao Yang, Soha Hassoun

Keywords: link prediction, few-shot learning, domain adaptation

Abstract: Real world link prediction problem often deals with data coming from multiple imbalanced domains. Similar problems in computer vision are often referred to as Few-Shot Learning (FSL) problems. However, for graph link prediction, this problem has rarely been addressed and explored. In this work, we propose an adversarial training based modification to the current state-of-the-arts link prediction method to solve this problem. We introduce a domain discriminator on pairs of graph-level embedding. We then use the discriminator to improve the model in an adversarial way, such that the graph embeddings generated by the model are domain agnostic. We test our proposal on 3 benchmark datasets. Our results demonstrate that, when domain differences exist, our method creates better graph embeddings that are more evenly distributed across domains and generates better prediction outcomes.

One-sentence Summary: Few-shot link prediction with multiple imbalanced domains could be improved by building domain agnostic embedding through adversarial training.