Path-aware Siamese Graph Neural Network for Link Prediction

TL;DR

This paper introduces a Path-aware Siamese Graph Neural Network (PSG) that leverages node and edge features, neighborhood structures, and relay paths, combined with self-supervised contrastive learning, to improve link prediction accuracy.

Contribution

The paper presents a novel PSG model with a multi-task GNN framework and contrastive learning for enhanced link prediction, outperforming existing methods on benchmark datasets.

Findings

Achieved top 1 performance on ogbl-ddi dataset.

Achieved top 3 performance on ogbl-collab dataset.

Demonstrated the effectiveness of path-aware and contrastive learning components.

Abstract

In this paper, we propose a Path-aware Siamese Graph neural network(PSG) for link prediction tasks. First, PSG captures both nodes and edge features for given two nodes, namely the structure information of k-neighborhoods and relay paths information of the nodes. Furthermore, a novel multi-task GNN framework with self-supervised contrastive learning is proposed for differentiation of positive links and negative links while content and behavior of nodes can be captured simultaneously. We evaluate the proposed algorithm PSG on two link property prediction datasets, ogbl-ddi and ogbl-collab. PSG achieves top 1 performance on ogbl-ddi until submission and top 3 performance on ogbl-collab. The experimental results verify the superiority of our proposed PSG