Hyperplane Knowledge Graph Embedding with Path Neighborhoods and Mapping Properties.

Knowledge representation learning(KRL) is significant for the knowledge graph completion(or link prediction) task, it aims to project entities and relations to a low-dimensional vector space. There are two topics have been widely studied in KRL: one is the ability of the model to handle complex relations(i.e., N-to-1, 1-to-N and N-to-N), and the other is whether the model integrates multi-source information. However, the existing methods rarely consider both topics. To mitigate this problem, this paper proposes TransPMH, a hyperplane knowledge graph embedding model with path neighborhoods and mapping properties, which models the relation as a hyperplane. Besides, this paper introduces path neighborhoods as the multi-source information to improve the model’s knowledge representation ability, and also introduces the relational mapping properties to enhance the model’s ability to handle complex relations. We conducted extensive experiments for link prediction and triplet classification on benchmark datasets like WordNet and Freebase. Experimental results show that the proposed method achieves significant improvements in multiple evaluation metrics.