Leap: Inductive Link Prediction via Learnable TopologyAugmentation

2025-03-05

AI Summary

To address the lack of structural context for newly introduced nodes in inductive link prediction, this paper proposes the first end-to-end learnable topological enhancement paradigm. The method integrates graph neural networks (GNNs), a differentiable graph generation module, and a meta-learning-driven neighborhood reweighting mechanism to dynamically optimize the topological neighborhood structure of unseen nodes, enabling structural-aware cross-graph generalization without retraining on unknown graphs. Compared to existing approaches—such as MLPs or static GNNs—our framework significantly improves inductive generalization: it achieves up to 12.6% higher accuracy across multiple standard benchmarks and improves inference efficiency by 40%. Crucially, it is the first approach to jointly model dynamic topology editing and cross-graph transfer, unifying structural adaptation and generalization in a single learnable framework.

Technology Category

• Machine Learning: Graph-based Machine Learning
• Search and Optimization: Learning to Search
• Data Mining & Knowledge Management: Graph Mining, Social Network Analysis & Community

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Problem

• Inductive link prediction for new nodes in graphs
• Capturing structural signals in graph machine learning
• Improving expressivity of node representations in GNNs

Innovation

• LEAP: inductive link prediction via learnable augmentation
• Combines structure and node features for expressivity
• Outperforms SOTA methods by up to 22% AUC