HIAE: Hyper-Relational Interaction Aware Embedding for Link Prediction

Abstract

Hyper-relational knowledge graph contains the main triple and additional information(qualifiers). The additional information can assist the main triple to predict missing entities(relations). In the previous methods, people usually combine the main triple and additional information through special methods to predict the missing entities(relations). However, not all additional information is equally essential for predicting missing entities(relations). Some additional information plays a more critical role in predicting missing entities (relations), so how to extract and use important additional information effectively in the link prediction (LP) task is very important. This paper proposes a novel Hyper-relational Interaction Aware Embedding (HIAE) model. Specifically, HIAE learns semantic features between the additional information and fuses the important additional information into the main triple through the attention mechanism to complete the prediction task. We have conducted many experiments to show that HIAE significantly outperforms the state-of-the-art models.