LAUSR

Knowledge graphs (KGs) play an important role in many real-world applications like information retrieval, question answering, relation extraction, etc. To reveal implicit knowledge from a knowledge graph (KG), viz. knowledge graph completion (KGC), is a crucial task for the downstream applications based on KG. For this purpose various embedding-based approaches have been proposed recently. This paper proposes a new approach named HRESCAL to KGC. It extends the well-known embedding-based approach RESCAL by introducing Hamming distance-based encoder to capture implicit multihop and partial inverse relation features in a KG. Experimental results on widely used KGC benchmarks show that the new approach achieves state-of-the-art or is competitive AUC performance.