Knowledge Graph Embedding by Learning to Connect Entity with Relation

Zichao Huang, Bo Li, Jian Yin

Abstract: Knowledge graph embedding aims to learn low-dimensional embedding vector representations for entities and relations, which can be used in further machine learning tasks. However, previous knowledge graph embedding models perform poorly when dealing with unbalanced relations which occupy a large proportion in knowledge graphs. In addition, modeling connections between entities and relations accurately is still a big challenge. In this paper, we propose a novel knowledge graph embedding model called ConnectER. It can solve the above problems through a “Connection-Classification” architecture. Experiment results show consistent improvements compared with state-of-the-art baselines.