Directed Link Prediction using GNN with Local and Global Feature Fusion

2025-06-25

AI Summary

To address the insufficient joint modeling of local neighborhood information and global topological structure in directed graph link prediction, this paper proposes a hybrid feature learning framework integrating node embeddings with directed community structures. Methodologically, it introduces: (1) a directed line graph transformation that maps directed edges to nodes, explicitly capturing inter-edge dependencies; (2) a dual-channel GNN architecture that concurrently aggregates local neighborhood messages and global community-aware signals; and (3) contrastive learning to enhance the discriminability of edge representations. Theoretical analysis demonstrates that the line graph transformation improves link predictability. Extensive experiments on multiple real-world directed graph benchmarks show that the model achieves significant performance gains over state-of-the-art methods using only 30%–60% of training data, validating its effectiveness, generalizability, and robustness.

Technology Category

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

Application Category

• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning 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

Abstract

Link prediction is a classical problem in graph analysis with many practical applications. For directed graphs, recently developed deep learning approaches typically analyze node similarities through contrastive learning and aggregate neighborhood information through graph convolutions. In this work, we propose a novel graph neural network (GNN) framework to fuse feature embedding with community information. We theoretically demonstrate that such hybrid features can improve the performance of directed link prediction. To utilize such features efficiently, we also propose an approach to transform input graphs into directed line graphs so that nodes in the transformed graph can aggregate more information during graph convolutions. Experiments on benchmark datasets show that our approach outperforms the state-of-the-art in most cases when 30%, 40%, 50%, and 60% of the connected links are used as training data, respectively.

Problem

• Improving directed link prediction using GNNs
• Fusing local and global graph features effectively
• Enhancing performance via directed line graph transformation

Innovation

• Fuses feature embedding with community information
• Transforms graphs into directed line graphs
• Uses hybrid features to improve link prediction