Mini-Batch Class Composition Bias in Link Prediction

TL;DR

This paper reveals that common link prediction models can exploit batch-normalization to learn trivial heuristics, which overstates their true ability to learn general graph representations.

Contribution

It uncovers a batch-normalization induced bias in link prediction models and demonstrates how correcting this bias improves the alignment with meaningful graph features.

Findings

Popular link prediction models can learn trivial heuristics due to batch-normalization.

Correcting for this bias increases the network's focus on node-class relevant features.

Standard training may overestimate the model's ability to learn general graph representations.

Abstract

Prior work on node classification has shown that Graph Neural Networks (GNNs) can learn representations that transfer across graphs, when underlying graph properties are shared. For a fixed graph, one would then expect GNNs trained for link prediction to learn a representation consistent with that learnt for node classification. We show this intuition does not hold in the general case. Instead, we find popular link prediction models can learn a trivial mini-batch dependent heuristic, enabled by batch-normalisation layers, to solve the edge classification task. When correcting for this, we observe increased alignment of the network representation with node-class relevant features, suggesting the network has learnt a graph representation that better aligns with the underlying graph's properties. Our findings suggest that standard link prediction training may be leading us to overestimate link predictors' ability to learn a generalised representation of a graph that is consistent across tasks.