Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Leap: Inductive Link Prediction via Learnable TopologyAugmentation

Ahmed E. Samy, Zekarias T. Kefato, Sarunas Girdzijauskas

TL;DR

LEAP tackles inductive link prediction by injecting topology information for new nodes through learnable anchor-based augmentations. It merges an MLP-Linker with a GNN operating on an augmented graph, enabling simultaneous use of node features and graph structure for unseen nodes. Across seven real-world graphs, LEAP consistently outperforms state-of-the-art baselines in both inductive and transductive settings, with notable gains in AUC and AP, especially on heterogeneous graphs. The approach offers a scalable, expressive framework for inductive graph learning and suggests future extensions to knowledge graphs and smarter anchor sampling strategies.

Abstract

Link prediction is a crucial task in many downstream applications of graph machine learning. To this end, Graph Neural Network (GNN) is a widely used technique for link prediction, mainly in transductive settings, where the goal is to predict missing links between existing nodes. However, many real-life applications require an inductive setting that accommodates for new nodes, coming into an existing graph. Thus, recently inductive link prediction has attracted considerable attention, and a multi-layer perceptron (MLP) is the popular choice of most studies to learn node representations. However, these approaches have limited expressivity and do not fully capture the graph's structural signal. Therefore, in this work we propose LEAP, an inductive link prediction method based on LEArnable toPology augmentation. Unlike previous methods, LEAP models the inductive bias from both the structure and node features, and hence is more expressive. To the best of our knowledge, this is the first attempt to provide structural contexts for new nodes via learnable augmentation in inductive settings. Extensive experiments on seven real-world homogeneous and heterogeneous graphs demonstrates that LEAP significantly surpasses SOTA methods. The improvements are up to 22\% and 17\% in terms of AUC and average precision, respectively. The code and datasets are available on GitHub (https://github.com/AhmedESamy/LEAP/)

Leap: Inductive Link Prediction via Learnable TopologyAugmentation

TL;DR

LEAP tackles inductive link prediction by injecting topology information for new nodes through learnable anchor-based augmentations. It merges an MLP-Linker with a GNN operating on an augmented graph, enabling simultaneous use of node features and graph structure for unseen nodes. Across seven real-world graphs, LEAP consistently outperforms state-of-the-art baselines in both inductive and transductive settings, with notable gains in AUC and AP, especially on heterogeneous graphs. The approach offers a scalable, expressive framework for inductive graph learning and suggests future extensions to knowledge graphs and smarter anchor sampling strategies.

Abstract

Link prediction is a crucial task in many downstream applications of graph machine learning. To this end, Graph Neural Network (GNN) is a widely used technique for link prediction, mainly in transductive settings, where the goal is to predict missing links between existing nodes. However, many real-life applications require an inductive setting that accommodates for new nodes, coming into an existing graph. Thus, recently inductive link prediction has attracted considerable attention, and a multi-layer perceptron (MLP) is the popular choice of most studies to learn node representations. However, these approaches have limited expressivity and do not fully capture the graph's structural signal. Therefore, in this work we propose LEAP, an inductive link prediction method based on LEArnable toPology augmentation. Unlike previous methods, LEAP models the inductive bias from both the structure and node features, and hence is more expressive. To the best of our knowledge, this is the first attempt to provide structural contexts for new nodes via learnable augmentation in inductive settings. Extensive experiments on seven real-world homogeneous and heterogeneous graphs demonstrates that LEAP significantly surpasses SOTA methods. The improvements are up to 22\% and 17\% in terms of AUC and average precision, respectively. The code and datasets are available on GitHub (https://github.com/AhmedESamy/LEAP/)

Paper Structure

This paper contains 18 sections, 9 equations, 2 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Preliminaries
  4. LEAP
  5. Anchor Selection
  6. Encoding and Decoding
  7. Training
  8. Experiments
  9. Datasets
  10. Baseline Methods
  11. Experimental Setup
  12. Inductive Link Prediction
  13. Transductive Link Prediction
  14. How does selecting the anchors affect LEAP?
  15. Hyperparameter sensitivity
  16. ...and 3 more sections

Figures (2)

  • Figure 1: An overview of LEAP; anchors per community are sampled (Fig1(a)). The red edges (Fig1(b)) signify the topology augmentation as predicted by the MLP. Each augmented/predicted edge is weighted by an activation from the final layer of the MLP and links a given newcomer node with an anchor node. The green edge is the inductive link to predict.
  • Figure 2: Anchor & Input Ablations