Meta-Analogy Learning Based on Dynamic Graph Neural Networks for Inductive Knowledge Graph Link Prediction

Abstract

For inductive link prediction in knowledge graphs, we address the problem by considering bridging links (i.e., links connecting discrete graphs). Although current research overcomes the traditional graph topological constraints, it tends to ignore the dynamic interactions between relations and entities as well as the capture of global information. The problem of discrete and small amounts of information in real-world knowledge graphs makes it difficult for existing methods to effectively integrate global and local information and to model complex relationships between entities. To address these issues, we propose a novel meta-analogy learning framework, Ank-motor, which integrates autonomously designed dynamic graph neural networks with analogical reasoning. The dynamic graph neural network models interactions and captures global information, while the analogical inference layer integrates entity, relation, and triple-layer information to capture local semantic details. In addition, meta-learning techniques are utilized to deal with problems with small amounts of data and to enhance the model’s ability to generalize to new tasks. Numerous experiments show that Ank-motor significantly outperforms existing models on multiple benchmark datasets.