Enhancing knowledge graph embedding by composite neighbors for link prediction

Knowledge graph embedding (KGE) aims to represent entities and relations in a low-dimensional continuous vector space. Recent KGE works focus on incorporating additional information, such as local neighbors and textual descriptions, to learn valuable representations. However, the non-uniformity and redundancy hinder the effectiveness of entity features from those information sources. In this paper, we propose a novel end-to-end framework, called composite neighborhood embedding (CoNE), utilizing composite neighbors to enhance the existing KGE methods. To ease past problems, the new composite neighbors are gathered from both entity descriptions and local neighbors. We design a novel Graph Memory Networks to extract entity features from composite neighbors, and fulfill the entity representation in the target KGE method. The experimental results show that CoNE effectively enhances three different KGE methods, TransE, ConvE, and RotatE, and achieves the state-of-the-art results on four real-world large datasets. Furthermore, our approach outperforms the recent text-enhanced models with fewer parameters and calculation. The source code of our work can be obtained from.