Semantic Interaction Matching Network for Few-shot Knowledge Graph Completion

The prosperity of knowledge graphs (KG), as well related downstream applications, have raised the urgent request graph completion techniques for fully supporting KG reasoning tasks, especially under circumstance training data scarcity. Though large efforts been made on solving this challenge via few-shot learning tools, they mainly focus simply aggregating entity neighbors to represent references, while enhancement from latent semantic correlation within has largely ignored. To that end, in paper, we propose a novel solution, named Semantic Interaction Matching network (SIM), which applies Transformer framework enhance representation with capturing interaction between neighbors. Specifically, first design entity-relation fusion module adaptively encode incorporating relation representation. Along line, layers are integrated capture neighbors, diversification support set. Finally, similarity score is attentively estimated attention mechanism. Extensive experiments two public benchmark datasets demonstrate our model outperforms variety state-of-the-art methods significant margin.