Knowledge Graph Embedding with Direct and Disentangled Neighborhood Representation Attention Network

Abstract

Knowledge Graph Completion (KGC) has become a focus of attention across the deep learning community owing to its excellent contribution to numerous downstream tasks. Despite the recent surge in KGC work, most existing KGC models deal with triples in Knowledge Graphs (KGs) independently, ignoring the inherent and valuable information from the neighborhoods around entities and the dynamic properties of entities in different link prediction tasks. We propose a novel Direct and Disentangled Neighborhood Representation Attention Network (DDNAN) for KGC, with an adaptive selector to decide what kind of neighborhood information should be aggregated to the current entity. With the assistance of the relation-aware attention aggregator, our model can exploit Knowledge Graphs (KGs) to generate dynamic representations of the given different scenarios and prediction tasks. Extensive experiments on public benchmark datasets have been conducted to validate the superiority of DDNAN over existing methods in terms of both accuracy and interpretability. The code is available at https://github.com/Ariel-Gao/DDNAN.