TuckerDNCaching: high-quality negative sampling with tucker decomposition

Abstract Knowledge Graph Embedding (KGE) translates entities and relations of knowledge graphs (KGs) into a low-dimensional vector space, enabling an efficient way predicting missing facts. Generally, KGE models are trained with positive negative examples, discriminating positives against negatives. Nevertheless, KGs contain only facts; training requires generating negatives from non-observed ones in KGs, referred to as sampling. Since sensitive inputs, sampling becomes crucial, the quality critical training. Generative adversarial networks (GAN) self-adversarial methods have recently been utilized address vanishing gradients observed early methods. However, they introduce problem false high probability. In this paper, we extend idea reducing by adopting Tucker decomposition representation, i.e., TuckerDNCaching, enhance semantic soundness latent among introducing relation feature space. TuckerDNCaching ensures generated samples, experimental results reflect that our proposed method outperforms existing state-of-the-art